generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Revert "Merge pull request #836 from jet47:gpu-modules"

Browse Source 
This reverts commit fba72cb60d, reversing
changes made to 02131ffb62.
This commit is contained in:
Andrey Kamaev
2013-04-18 15:03:50 +04:00
parent fba72cb60d
commit 416fb50594
472 changed files with 22945 additions and 29803 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
modules/gpuimgproc/CMakeLists.txt
View File

@@ -1,9 +0,0 @@
if(ANDROID OR IOS)
ocv_module_disable(gpuimgproc)
endif()
set(the_description "GPU-accelerated Image Processing")
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4127 /wd4100 /wd4324 /wd4512 -Wundef -Wmissing-declarations -Wshadow -Wunused-parameter)
ocv_define_module(gpuimgproc opencv_imgproc opencv_gpufilters OPTIONAL opencv_gpuarithm)

74
modules/gpuimgproc/doc/color.rst
View File

@@ -1,74 +0,0 @@
Color space processing
======================
.. highlight:: cpp
gpu::cvtColor
-----------------
Converts an image from one color space to another.
.. ocv:function:: void gpu::cvtColor(const GpuMat& src, GpuMat& dst, int code, int dcn = 0, Stream& stream = Stream::Null())
:param src: Source image with ``CV_8U`` , ``CV_16U`` , or ``CV_32F`` depth and 1, 3, or 4 channels.
:param dst: Destination image with the same size and depth as ``src`` .
:param code: Color space conversion code. For details, see :ocv:func:`cvtColor` . Conversion to/from Luv and Bayer color spaces is not supported.
:param dcn: Number of channels in the destination image. If the parameter is 0, the number of the channels is derived automatically from ``src`` and the ``code`` .
:param stream: Stream for the asynchronous version.
3-channel color spaces (like ``HSV``, ``XYZ``, and so on) can be stored in a 4-channel image for better performance.
.. seealso:: :ocv:func:`cvtColor`
gpu::swapChannels
-----------------
Exchanges the color channels of an image in-place.
.. ocv:function:: void gpu::swapChannels(GpuMat& image, const int dstOrder[4], Stream& stream = Stream::Null())
:param image: Source image. Supports only ``CV_8UC4`` type.
:param dstOrder: Integer array describing how channel values are permutated. The n-th entry of the array contains the number of the channel that is stored in the n-th channel of the output image. E.g. Given an RGBA image, aDstOrder = [3,2,1,0] converts this to ABGR channel order.
:param stream: Stream for the asynchronous version.
The methods support arbitrary permutations of the original channels, including replication.
gpu::alphaComp
-------------------
Composites two images using alpha opacity values contained in each image.
.. ocv:function:: void gpu::alphaComp(const GpuMat& img1, const GpuMat& img2, GpuMat& dst, int alpha_op, Stream& stream = Stream::Null())
:param img1: First image. Supports ``CV_8UC4`` , ``CV_16UC4`` , ``CV_32SC4`` and ``CV_32FC4`` types.
:param img2: Second image. Must have the same size and the same type as ``img1`` .
:param dst: Destination image.
:param alpha_op: Flag specifying the alpha-blending operation:
* **ALPHA_OVER**
* **ALPHA_IN**
* **ALPHA_OUT**
* **ALPHA_ATOP**
* **ALPHA_XOR**
* **ALPHA_PLUS**
* **ALPHA_OVER_PREMUL**
* **ALPHA_IN_PREMUL**
* **ALPHA_OUT_PREMUL**
* **ALPHA_ATOP_PREMUL**
* **ALPHA_XOR_PREMUL**
* **ALPHA_PLUS_PREMUL**
* **ALPHA_PREMUL**
:param stream: Stream for the asynchronous version.

81
modules/gpuimgproc/doc/feature_detection.rst
View File

@@ -1,81 +0,0 @@
Feature Detection
=================
.. highlight:: cpp
gpu::cornerHarris
---------------------
Computes the Harris cornerness criteria at each image pixel.
.. ocv:function:: void gpu::cornerHarris(const GpuMat& src, GpuMat& dst, int blockSize, int ksize, double k, int borderType=BORDER_REFLECT101)
:param src: Source image. Only ``CV_8UC1`` and ``CV_32FC1`` images are supported for now.
:param dst: Destination image containing cornerness values. It has the same size as ``src`` and ``CV_32FC1`` type.
:param blockSize: Neighborhood size.
:param ksize: Aperture parameter for the Sobel operator.
:param k: Harris detector free parameter.
:param borderType: Pixel extrapolation method. Only ``BORDER_REFLECT101`` and ``BORDER_REPLICATE`` are supported for now.
.. seealso:: :ocv:func:`cornerHarris`
gpu::cornerMinEigenVal
--------------------------
Computes the minimum eigen value of a 2x2 derivative covariation matrix at each pixel (the cornerness criteria).
.. ocv:function:: void gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, int blockSize, int ksize, int borderType=BORDER_REFLECT101)
.. ocv:function:: void gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, int blockSize, int ksize, int borderType=BORDER_REFLECT101)
.. ocv:function:: void gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, GpuMat& buf, int blockSize, int ksize, int borderType=BORDER_REFLECT101, Stream& stream = Stream::Null())
:param src: Source image. Only ``CV_8UC1`` and ``CV_32FC1`` images are supported for now.
:param dst: Destination image containing cornerness values. The size is the same. The type is ``CV_32FC1`` .
:param blockSize: Neighborhood size.
:param ksize: Aperture parameter for the Sobel operator.
:param borderType: Pixel extrapolation method. Only ``BORDER_REFLECT101`` and ``BORDER_REPLICATE`` are supported for now.
.. seealso:: :ocv:func:`cornerMinEigenVal`
gpu::GoodFeaturesToTrackDetector_GPU
------------------------------------
.. ocv:class:: gpu::GoodFeaturesToTrackDetector_GPU
Class used for strong corners detection on an image. ::
class GoodFeaturesToTrackDetector_GPU
{
public:
explicit GoodFeaturesToTrackDetector_GPU(int maxCorners_ = 1000, double qualityLevel_ = 0.01, double minDistance_ = 0.0,
int blockSize_ = 3, bool useHarrisDetector_ = false, double harrisK_ = 0.04);
void operator ()(const GpuMat& image, GpuMat& corners, const GpuMat& mask = GpuMat());
int maxCorners;
double qualityLevel;
double minDistance;
int blockSize;
bool useHarrisDetector;
double harrisK;
void releaseMemory();
};
The class finds the most prominent corners in the image.
.. seealso:: :ocv:func:`goodFeaturesToTrack`

12
modules/gpuimgproc/doc/gpuimgproc.rst
View File

@@ -1,12 +0,0 @@
********************************************
gpuimgproc. GPU-accelerated Image Processing
********************************************
.. toctree::
:maxdepth: 1
color
histogram
hough
feature_detection
imgproc

104
modules/gpuimgproc/doc/histogram.rst
View File

@@ -1,104 +0,0 @@
Histogram Calculation
=====================
.. highlight:: cpp
gpu::evenLevels
-------------------
Computes levels with even distribution.
.. ocv:function:: void gpu::evenLevels(GpuMat& levels, int nLevels, int lowerLevel, int upperLevel)
:param levels: Destination array. ``levels`` has 1 row, ``nLevels`` columns, and the ``CV_32SC1`` type.
:param nLevels: Number of computed levels. ``nLevels`` must be at least 2.
:param lowerLevel: Lower boundary value of the lowest level.
:param upperLevel: Upper boundary value of the greatest level.
gpu::histEven
-----------------
Calculates a histogram with evenly distributed bins.
.. ocv:function:: void gpu::histEven(const GpuMat& src, GpuMat& hist, int histSize, int lowerLevel, int upperLevel, Stream& stream = Stream::Null())
.. ocv:function:: void gpu::histEven(const GpuMat& src, GpuMat& hist, GpuMat& buf, int histSize, int lowerLevel, int upperLevel, Stream& stream = Stream::Null())
.. ocv:function:: void gpu::histEven( const GpuMat& src, GpuMat hist[4], int histSize[4], int lowerLevel[4], int upperLevel[4], Stream& stream=Stream::Null() )
.. ocv:function:: void gpu::histEven( const GpuMat& src, GpuMat hist[4], GpuMat& buf, int histSize[4], int lowerLevel[4], int upperLevel[4], Stream& stream=Stream::Null() )
:param src: Source image. ``CV_8U``, ``CV_16U``, or ``CV_16S`` depth and 1 or 4 channels are supported. For a four-channel image, all channels are processed separately.
:param hist: Destination histogram with one row, ``histSize`` columns, and the ``CV_32S`` type.
:param histSize: Size of the histogram.
:param lowerLevel: Lower boundary of lowest-level bin.
:param upperLevel: Upper boundary of highest-level bin.
:param buf: Optional buffer to avoid extra memory allocations (for many calls with the same sizes).
:param stream: Stream for the asynchronous version.
gpu::histRange
------------------
Calculates a histogram with bins determined by the ``levels`` array.
.. ocv:function:: void gpu::histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels, Stream& stream = Stream::Null())
.. ocv:function:: void gpu::histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels, GpuMat& buf, Stream& stream = Stream::Null())
:param src: Source image. ``CV_8U`` , ``CV_16U`` , or ``CV_16S`` depth and 1 or 4 channels are supported. For a four-channel image, all channels are processed separately.
:param hist: Destination histogram with one row, ``(levels.cols-1)`` columns, and the ``CV_32SC1`` type.
:param levels: Number of levels in the histogram.
:param buf: Optional buffer to avoid extra memory allocations (for many calls with the same sizes).
:param stream: Stream for the asynchronous version.
gpu::calcHist
------------------
Calculates histogram for one channel 8-bit image.
.. ocv:function:: void gpu::calcHist(const GpuMat& src, GpuMat& hist, Stream& stream = Stream::Null())
:param src: Source image.
:param hist: Destination histogram with one row, 256 columns, and the ``CV_32SC1`` type.
:param stream: Stream for the asynchronous version.
gpu::equalizeHist
------------------
Equalizes the histogram of a grayscale image.
.. ocv:function:: void gpu::equalizeHist(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null())
.. ocv:function:: void gpu::equalizeHist(const GpuMat& src, GpuMat& dst, GpuMat& hist, GpuMat& buf, Stream& stream = Stream::Null())
:param src: Source image.
:param dst: Destination image.
:param hist: Destination histogram with one row, 256 columns, and the ``CV_32SC1`` type.
:param buf: Optional buffer to avoid extra memory allocations (for many calls with the same sizes).
:param stream: Stream for the asynchronous version.
.. seealso:: :ocv:func:`equalizeHist`

96
modules/gpuimgproc/doc/hough.rst
View File

@@ -1,96 +0,0 @@
Hough Transform
===============
.. highlight:: cpp
gpu::HoughLines
---------------
Finds lines in a binary image using the classical Hough transform.
.. ocv:function:: void gpu::HoughLines(const GpuMat& src, GpuMat& lines, float rho, float theta, int threshold, bool doSort = false, int maxLines = 4096)
.. ocv:function:: void gpu::HoughLines(const GpuMat& src, GpuMat& lines, HoughLinesBuf& buf, float rho, float theta, int threshold, bool doSort = false, int maxLines = 4096)
:param src: 8-bit, single-channel binary source image.
:param lines: Output vector of lines. Each line is represented by a two-element vector :math:`(\rho, \theta)` . :math:`\rho` is the distance from the coordinate origin :math:`(0,0)` (top-left corner of the image). :math:`\theta` is the line rotation angle in radians ( :math:`0 \sim \textrm{vertical line}, \pi/2 \sim \textrm{horizontal line}` ).
:param rho: Distance resolution of the accumulator in pixels.
:param theta: Angle resolution of the accumulator in radians.
:param threshold: Accumulator threshold parameter. Only those lines are returned that get enough votes ( :math:`>\texttt{threshold}` ).
:param doSort: Performs lines sort by votes.
:param maxLines: Maximum number of output lines.
:param buf: Optional buffer to avoid extra memory allocations (for many calls with the same sizes).
.. seealso:: :ocv:func:`HoughLines`
gpu::HoughLinesDownload
-----------------------
Downloads results from :ocv:func:`gpu::HoughLines` to host memory.
.. ocv:function:: void gpu::HoughLinesDownload(const GpuMat& d_lines, OutputArray h_lines, OutputArray h_votes = noArray())
:param d_lines: Result of :ocv:func:`gpu::HoughLines` .
:param h_lines: Output host array.
:param h_votes: Optional output array for line's votes.
.. seealso:: :ocv:func:`gpu::HoughLines`
gpu::HoughCircles
-----------------
Finds circles in a grayscale image using the Hough transform.
.. ocv:function:: void gpu::HoughCircles(const GpuMat& src, GpuMat& circles, int method, float dp, float minDist, int cannyThreshold, int votesThreshold, int minRadius, int maxRadius, int maxCircles = 4096)
.. ocv:function:: void gpu::HoughCircles(const GpuMat& src, GpuMat& circles, HoughCirclesBuf& buf, int method, float dp, float minDist, int cannyThreshold, int votesThreshold, int minRadius, int maxRadius, int maxCircles = 4096)
:param src: 8-bit, single-channel grayscale input image.
:param circles: Output vector of found circles. Each vector is encoded as a 3-element floating-point vector :math:`(x, y, radius)` .
:param method: Detection method to use. Currently, the only implemented method is ``CV_HOUGH_GRADIENT`` , which is basically *21HT* , described in [Yuen90]_.
:param dp: Inverse ratio of the accumulator resolution to the image resolution. For example, if ``dp=1`` , the accumulator has the same resolution as the input image. If ``dp=2`` , the accumulator has half as big width and height.
:param minDist: Minimum distance between the centers of the detected circles. If the parameter is too small, multiple neighbor circles may be falsely detected in addition to a true one. If it is too large, some circles may be missed.
:param cannyThreshold: The higher threshold of the two passed to the :ocv:func:`gpu::Canny` edge detector (the lower one is twice smaller).
:param votesThreshold: The accumulator threshold for the circle centers at the detection stage. The smaller it is, the more false circles may be detected.
:param minRadius: Minimum circle radius.
:param maxRadius: Maximum circle radius.
:param maxCircles: Maximum number of output circles.
:param buf: Optional buffer to avoid extra memory allocations (for many calls with the same sizes).
.. seealso:: :ocv:func:`HoughCircles`
gpu::HoughCirclesDownload
-------------------------
Downloads results from :ocv:func:`gpu::HoughCircles` to host memory.
.. ocv:function:: void gpu::HoughCirclesDownload(const GpuMat& d_circles, OutputArray h_circles)
:param d_circles: Result of :ocv:func:`gpu::HoughCircles` .
:param h_circles: Output host array.
.. seealso:: :ocv:func:`gpu::HoughCircles`

203
modules/gpuimgproc/doc/imgproc.rst
View File

@@ -1,203 +0,0 @@
Image Processing
================
.. highlight:: cpp
gpu::meanShiftFiltering
---------------------------
Performs mean-shift filtering for each point of the source image.
.. ocv:function:: void gpu::meanShiftFiltering( const GpuMat& src, GpuMat& dst, int sp, int sr, TermCriteria criteria=TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1), Stream& stream=Stream::Null() )
:param src: Source image. Only ``CV_8UC4`` images are supported for now.
:param dst: Destination image containing the color of mapped points. It has the same size and type as ``src`` .
:param sp: Spatial window radius.
:param sr: Color window radius.
:param criteria: Termination criteria. See :ocv:class:`TermCriteria`.
It maps each point of the source image into another point. As a result, you have a new color and new position of each point.
gpu::meanShiftProc
----------------------
Performs a mean-shift procedure and stores information about processed points (their colors and positions) in two images.
.. ocv:function:: void gpu::meanShiftProc( const GpuMat& src, GpuMat& dstr, GpuMat& dstsp, int sp, int sr, TermCriteria criteria=TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1), Stream& stream=Stream::Null() )
:param src: Source image. Only ``CV_8UC4`` images are supported for now.
:param dstr: Destination image containing the color of mapped points. The size and type is the same as ``src`` .
:param dstsp: Destination image containing the position of mapped points. The size is the same as ``src`` size. The type is ``CV_16SC2`` .
:param sp: Spatial window radius.
:param sr: Color window radius.
:param criteria: Termination criteria. See :ocv:class:`TermCriteria`.
.. seealso:: :ocv:func:`gpu::meanShiftFiltering`
gpu::meanShiftSegmentation
------------------------------
Performs a mean-shift segmentation of the source image and eliminates small segments.
.. ocv:function:: void gpu::meanShiftSegmentation(const GpuMat& src, Mat& dst, int sp, int sr, int minsize, TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1))
:param src: Source image. Only ``CV_8UC4`` images are supported for now.
:param dst: Segmented image with the same size and type as ``src`` .
:param sp: Spatial window radius.
:param sr: Color window radius.
:param minsize: Minimum segment size. Smaller segments are merged.
:param criteria: Termination criteria. See :ocv:class:`TermCriteria`.
gpu::MatchTemplateBuf
---------------------
.. ocv:struct:: gpu::MatchTemplateBuf
Class providing memory buffers for :ocv:func:`gpu::matchTemplate` function, plus it allows to adjust some specific parameters. ::
struct CV_EXPORTS MatchTemplateBuf
{
Size user_block_size;
GpuMat imagef, templf;
std::vector<GpuMat> images;
std::vector<GpuMat> image_sums;
std::vector<GpuMat> image_sqsums;
};
You can use field `user_block_size` to set specific block size for :ocv:func:`gpu::matchTemplate` function. If you leave its default value `Size(0,0)` then automatic estimation of block size will be used (which is optimized for speed). By varying `user_block_size` you can reduce memory requirements at the cost of speed.
gpu::matchTemplate
----------------------
Computes a proximity map for a raster template and an image where the template is searched for.
.. ocv:function:: void gpu::matchTemplate(const GpuMat& image, const GpuMat& templ, GpuMat& result, int method, Stream &stream = Stream::Null())
.. ocv:function:: void gpu::matchTemplate(const GpuMat& image, const GpuMat& templ, GpuMat& result, int method, MatchTemplateBuf &buf, Stream& stream = Stream::Null())
:param image: Source image. ``CV_32F`` and ``CV_8U`` depth images (1..4 channels) are supported for now.
:param templ: Template image with the size and type the same as ``image`` .
:param result: Map containing comparison results ( ``CV_32FC1`` ). If ``image`` is *W x H* and ``templ`` is *w x h*, then ``result`` must be *W-w+1 x H-h+1*.
:param method: Specifies the way to compare the template with the image.
:param buf: Optional buffer to avoid extra memory allocations and to adjust some specific parameters. See :ocv:struct:`gpu::MatchTemplateBuf`.
:param stream: Stream for the asynchronous version.
The following methods are supported for the ``CV_8U`` depth images for now:
* ``CV_TM_SQDIFF``
* ``CV_TM_SQDIFF_NORMED``
* ``CV_TM_CCORR``
* ``CV_TM_CCORR_NORMED``
* ``CV_TM_CCOEFF``
* ``CV_TM_CCOEFF_NORMED``
The following methods are supported for the ``CV_32F`` images for now:
* ``CV_TM_SQDIFF``
* ``CV_TM_CCORR``
.. seealso:: :ocv:func:`matchTemplate`
gpu::Canny
-------------------
Finds edges in an image using the [Canny86]_ algorithm.
.. ocv:function:: void gpu::Canny(const GpuMat& image, GpuMat& edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false)
.. ocv:function:: void gpu::Canny(const GpuMat& image, CannyBuf& buf, GpuMat& edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false)
.. ocv:function:: void gpu::Canny(const GpuMat& dx, const GpuMat& dy, GpuMat& edges, double low_thresh, double high_thresh, bool L2gradient = false)
.. ocv:function:: void gpu::Canny(const GpuMat& dx, const GpuMat& dy, CannyBuf& buf, GpuMat& edges, double low_thresh, double high_thresh, bool L2gradient = false)
:param image: Single-channel 8-bit input image.
:param dx: First derivative of image in the vertical direction. Support only ``CV_32S`` type.
:param dy: First derivative of image in the horizontal direction. Support only ``CV_32S`` type.
:param edges: Output edge map. It has the same size and type as ``image`` .
:param low_thresh: First threshold for the hysteresis procedure.
:param high_thresh: Second threshold for the hysteresis procedure.
:param apperture_size: Aperture size for the :ocv:func:`Sobel` operator.
:param L2gradient: Flag indicating whether a more accurate :math:`L_2` norm :math:`=\sqrt{(dI/dx)^2 + (dI/dy)^2}` should be used to compute the image gradient magnitude ( ``L2gradient=true`` ), or a faster default :math:`L_1` norm :math:`=|dI/dx|+|dI/dy|` is enough ( ``L2gradient=false`` ).
:param buf: Optional buffer to avoid extra memory allocations (for many calls with the same sizes).
.. seealso:: :ocv:func:`Canny`
gpu::bilateralFilter
--------------------
Performs bilateral filtering of passed image
.. ocv:function:: void gpu::bilateralFilter( const GpuMat& src, GpuMat& dst, int kernel_size, float sigma_color, float sigma_spatial, int borderMode=BORDER_DEFAULT, Stream& stream=Stream::Null() )
:param src: Source image. Supports only (channles != 2 && depth() != CV_8S && depth() != CV_32S && depth() != CV_64F).
:param dst: Destination imagwe.
:param kernel_size: Kernel window size.
:param sigma_color: Filter sigma in the color space.
:param sigma_spatial: Filter sigma in the coordinate space.
:param borderMode: Border type. See :ocv:func:`borderInterpolate` for details. ``BORDER_REFLECT101`` , ``BORDER_REPLICATE`` , ``BORDER_CONSTANT`` , ``BORDER_REFLECT`` and ``BORDER_WRAP`` are supported for now.
:param stream: Stream for the asynchronous version.
.. seealso::
:ocv:func:`bilateralFilter`
gpu::blendLinear
-------------------
Performs linear blending of two images.
.. ocv:function:: void gpu::blendLinear(const GpuMat& img1, const GpuMat& img2, const GpuMat& weights1, const GpuMat& weights2, GpuMat& result, Stream& stream = Stream::Null())
:param img1: First image. Supports only ``CV_8U`` and ``CV_32F`` depth.
:param img2: Second image. Must have the same size and the same type as ``img1`` .
:param weights1: Weights for first image. Must have tha same size as ``img1`` . Supports only ``CV_32F`` type.
:param weights2: Weights for second image. Must have tha same size as ``img2`` . Supports only ``CV_32F`` type.
:param result: Destination image.
:param stream: Stream for the asynchronous version.

339
modules/gpuimgproc/include/opencv2/gpuimgproc.hpp
View File

@@ -1,339 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_GPUIMGPROC_HPP__
#define __OPENCV_GPUIMGPROC_HPP__
#ifndef __cplusplus
# error gpuimgproc.hpp header must be compiled as C++
#endif
#include "opencv2/core/gpumat.hpp"
#include "opencv2/core/base.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/gpufilters.hpp"
namespace cv { namespace gpu {
/////////////////////////// Color Processing ///////////////////////////
//! converts image from one color space to another
CV_EXPORTS void cvtColor(const GpuMat& src, GpuMat& dst, int code, int dcn = 0, Stream& stream = Stream::Null());
enum
{
// Bayer Demosaicing (Malvar, He, and Cutler)
COLOR_BayerBG2BGR_MHT = 256,
COLOR_BayerGB2BGR_MHT = 257,
COLOR_BayerRG2BGR_MHT = 258,
COLOR_BayerGR2BGR_MHT = 259,
COLOR_BayerBG2RGB_MHT = COLOR_BayerRG2BGR_MHT,
COLOR_BayerGB2RGB_MHT = COLOR_BayerGR2BGR_MHT,
COLOR_BayerRG2RGB_MHT = COLOR_BayerBG2BGR_MHT,
COLOR_BayerGR2RGB_MHT = COLOR_BayerGB2BGR_MHT,
COLOR_BayerBG2GRAY_MHT = 260,
COLOR_BayerGB2GRAY_MHT = 261,
COLOR_BayerRG2GRAY_MHT = 262,
COLOR_BayerGR2GRAY_MHT = 263
};
CV_EXPORTS void demosaicing(const GpuMat& src, GpuMat& dst, int code, int dcn = -1, Stream& stream = Stream::Null());
//! swap channels
//! dstOrder - Integer array describing how channel values are permutated. The n-th entry
//! of the array contains the number of the channel that is stored in the n-th channel of
//! the output image. E.g. Given an RGBA image, aDstOrder = [3,2,1,0] converts this to ABGR
//! channel order.
CV_EXPORTS void swapChannels(GpuMat& image, const int dstOrder[4], Stream& stream = Stream::Null());
//! Routines for correcting image color gamma
CV_EXPORTS void gammaCorrection(const GpuMat& src, GpuMat& dst, bool forward = true, Stream& stream = Stream::Null());
enum { ALPHA_OVER, ALPHA_IN, ALPHA_OUT, ALPHA_ATOP, ALPHA_XOR, ALPHA_PLUS, ALPHA_OVER_PREMUL, ALPHA_IN_PREMUL, ALPHA_OUT_PREMUL,
ALPHA_ATOP_PREMUL, ALPHA_XOR_PREMUL, ALPHA_PLUS_PREMUL, ALPHA_PREMUL};
//! Composite two images using alpha opacity values contained in each image
//! Supports CV_8UC4, CV_16UC4, CV_32SC4 and CV_32FC4 types
CV_EXPORTS void alphaComp(const GpuMat& img1, const GpuMat& img2, GpuMat& dst, int alpha_op, Stream& stream = Stream::Null());
////////////////////////////// Histogram ///////////////////////////////
//! Compute levels with even distribution. levels will have 1 row and nLevels cols and CV_32SC1 type.
CV_EXPORTS void evenLevels(GpuMat& levels, int nLevels, int lowerLevel, int upperLevel);
//! Calculates histogram with evenly distributed bins for signle channel source.
//! Supports CV_8UC1, CV_16UC1 and CV_16SC1 source types.
//! Output hist will have one row and histSize cols and CV_32SC1 type.
CV_EXPORTS void histEven(const GpuMat& src, GpuMat& hist, int histSize, int lowerLevel, int upperLevel, Stream& stream = Stream::Null());
CV_EXPORTS void histEven(const GpuMat& src, GpuMat& hist, GpuMat& buf, int histSize, int lowerLevel, int upperLevel, Stream& stream = Stream::Null());
//! Calculates histogram with evenly distributed bins for four-channel source.
//! All channels of source are processed separately.
//! Supports CV_8UC4, CV_16UC4 and CV_16SC4 source types.
//! Output hist[i] will have one row and histSize[i] cols and CV_32SC1 type.
CV_EXPORTS void histEven(const GpuMat& src, GpuMat hist[4], int histSize[4], int lowerLevel[4], int upperLevel[4], Stream& stream = Stream::Null());
CV_EXPORTS void histEven(const GpuMat& src, GpuMat hist[4], GpuMat& buf, int histSize[4], int lowerLevel[4], int upperLevel[4], Stream& stream = Stream::Null());
//! Calculates histogram with bins determined by levels array.
//! levels must have one row and CV_32SC1 type if source has integer type or CV_32FC1 otherwise.
//! Supports CV_8UC1, CV_16UC1, CV_16SC1 and CV_32FC1 source types.
//! Output hist will have one row and (levels.cols-1) cols and CV_32SC1 type.
CV_EXPORTS void histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels, Stream& stream = Stream::Null());
CV_EXPORTS void histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels, GpuMat& buf, Stream& stream = Stream::Null());
//! Calculates histogram with bins determined by levels array.
//! All levels must have one row and CV_32SC1 type if source has integer type or CV_32FC1 otherwise.
//! All channels of source are processed separately.
//! Supports CV_8UC4, CV_16UC4, CV_16SC4 and CV_32FC4 source types.
//! Output hist[i] will have one row and (levels[i].cols-1) cols and CV_32SC1 type.
CV_EXPORTS void histRange(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4], Stream& stream = Stream::Null());
CV_EXPORTS void histRange(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4], GpuMat& buf, Stream& stream = Stream::Null());
//! Calculates histogram for 8u one channel image
//! Output hist will have one row, 256 cols and CV32SC1 type.
CV_EXPORTS void calcHist(const GpuMat& src, GpuMat& hist, Stream& stream = Stream::Null());
CV_EXPORTS void calcHist(const GpuMat& src, GpuMat& hist, GpuMat& buf, Stream& stream = Stream::Null());
//! normalizes the grayscale image brightness and contrast by normalizing its histogram
CV_EXPORTS void equalizeHist(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null());
CV_EXPORTS void equalizeHist(const GpuMat& src, GpuMat& dst, GpuMat& hist, Stream& stream = Stream::Null());
CV_EXPORTS void equalizeHist(const GpuMat& src, GpuMat& dst, GpuMat& hist, GpuMat& buf, Stream& stream = Stream::Null());
class CV_EXPORTS CLAHE : public cv::CLAHE
{
public:
using cv::CLAHE::apply;
virtual void apply(InputArray src, OutputArray dst, Stream& stream) = 0;
};
CV_EXPORTS Ptr<cv::gpu::CLAHE> createCLAHE(double clipLimit = 40.0, Size tileGridSize = Size(8, 8));
//////////////////////////////// Canny ////////////////////////////////
struct CV_EXPORTS CannyBuf
{
void create(const Size& image_size, int apperture_size = 3);
void release();
GpuMat dx, dy;
GpuMat mag;
GpuMat map;
GpuMat st1, st2;
Ptr<FilterEngine_GPU> filterDX, filterDY;
};
CV_EXPORTS void Canny(const GpuMat& image, GpuMat& edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
CV_EXPORTS void Canny(const GpuMat& image, CannyBuf& buf, GpuMat& edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false);
CV_EXPORTS void Canny(const GpuMat& dx, const GpuMat& dy, GpuMat& edges, double low_thresh, double high_thresh, bool L2gradient = false);
CV_EXPORTS void Canny(const GpuMat& dx, const GpuMat& dy, CannyBuf& buf, GpuMat& edges, double low_thresh, double high_thresh, bool L2gradient = false);
/////////////////////////// Hough Transform ////////////////////////////
struct HoughLinesBuf
{
GpuMat accum;
GpuMat list;
};
CV_EXPORTS void HoughLines(const GpuMat& src, GpuMat& lines, float rho, float theta, int threshold, bool doSort = false, int maxLines = 4096);
CV_EXPORTS void HoughLines(const GpuMat& src, GpuMat& lines, HoughLinesBuf& buf, float rho, float theta, int threshold, bool doSort = false, int maxLines = 4096);
CV_EXPORTS void HoughLinesDownload(const GpuMat& d_lines, OutputArray h_lines, OutputArray h_votes = noArray());
//! finds line segments in the black-n-white image using probabalistic Hough transform
CV_EXPORTS void HoughLinesP(const GpuMat& image, GpuMat& lines, HoughLinesBuf& buf, float rho, float theta, int minLineLength, int maxLineGap, int maxLines = 4096);
struct HoughCirclesBuf
{
GpuMat edges;
GpuMat accum;
GpuMat list;
CannyBuf cannyBuf;
};
CV_EXPORTS void HoughCircles(const GpuMat& src, GpuMat& circles, int method, float dp, float minDist, int cannyThreshold, int votesThreshold, int minRadius, int maxRadius, int maxCircles = 4096);
CV_EXPORTS void HoughCircles(const GpuMat& src, GpuMat& circles, HoughCirclesBuf& buf, int method, float dp, float minDist, int cannyThreshold, int votesThreshold, int minRadius, int maxRadius, int maxCircles = 4096);
CV_EXPORTS void HoughCirclesDownload(const GpuMat& d_circles, OutputArray h_circles);
//! finds arbitrary template in the grayscale image using Generalized Hough Transform
//! Ballard, D.H. (1981). Generalizing the Hough transform to detect arbitrary shapes. Pattern Recognition 13 (2): 111-122.
//! Guil, N., González-Linares, J.M. and Zapata, E.L. (1999). Bidimensional shape detection using an invariant approach. Pattern Recognition 32 (6): 1025-1038.
class CV_EXPORTS GeneralizedHough_GPU : public cv::Algorithm
{
public:
static Ptr<GeneralizedHough_GPU> create(int method);
virtual ~GeneralizedHough_GPU();
//! set template to search
void setTemplate(const GpuMat& templ, int cannyThreshold = 100, Point templCenter = Point(-1, -1));
void setTemplate(const GpuMat& edges, const GpuMat& dx, const GpuMat& dy, Point templCenter = Point(-1, -1));
//! find template on image
void detect(const GpuMat& image, GpuMat& positions, int cannyThreshold = 100);
void detect(const GpuMat& edges, const GpuMat& dx, const GpuMat& dy, GpuMat& positions);
void download(const GpuMat& d_positions, OutputArray h_positions, OutputArray h_votes = noArray());
void release();
protected:
virtual void setTemplateImpl(const GpuMat& edges, const GpuMat& dx, const GpuMat& dy, Point templCenter) = 0;
virtual void detectImpl(const GpuMat& edges, const GpuMat& dx, const GpuMat& dy, GpuMat& positions) = 0;
virtual void releaseImpl() = 0;
private:
GpuMat edges_;
CannyBuf cannyBuf_;
};
////////////////////////// Corners Detection ///////////////////////////
//! computes Harris cornerness criteria at each image pixel
CV_EXPORTS void cornerHarris(const GpuMat& src, GpuMat& dst, int blockSize, int ksize, double k, int borderType = BORDER_REFLECT101);
CV_EXPORTS void cornerHarris(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, int blockSize, int ksize, double k, int borderType = BORDER_REFLECT101);
CV_EXPORTS void cornerHarris(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, GpuMat& buf, int blockSize, int ksize, double k,
int borderType = BORDER_REFLECT101, Stream& stream = Stream::Null());
//! computes minimum eigen value of 2x2 derivative covariation matrix at each pixel - the cornerness criteria
CV_EXPORTS void cornerMinEigenVal(const GpuMat& src, GpuMat& dst, int blockSize, int ksize, int borderType=BORDER_REFLECT101);
CV_EXPORTS void cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, int blockSize, int ksize, int borderType=BORDER_REFLECT101);
CV_EXPORTS void cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, GpuMat& buf, int blockSize, int ksize,
int borderType=BORDER_REFLECT101, Stream& stream = Stream::Null());
////////////////////////// Feature Detection ///////////////////////////
class CV_EXPORTS GoodFeaturesToTrackDetector_GPU
{
public:
explicit GoodFeaturesToTrackDetector_GPU(int maxCorners = 1000, double qualityLevel = 0.01, double minDistance = 0.0,
int blockSize = 3, bool useHarrisDetector = false, double harrisK = 0.04);
//! return 1 rows matrix with CV_32FC2 type
void operator ()(const GpuMat& image, GpuMat& corners, const GpuMat& mask = GpuMat());
int maxCorners;
double qualityLevel;
double minDistance;
int blockSize;
bool useHarrisDetector;
double harrisK;
void releaseMemory()
{
Dx_.release();
Dy_.release();
buf_.release();
eig_.release();
minMaxbuf_.release();
tmpCorners_.release();
}
private:
GpuMat Dx_;
GpuMat Dy_;
GpuMat buf_;
GpuMat eig_;
GpuMat minMaxbuf_;
GpuMat tmpCorners_;
};
inline GoodFeaturesToTrackDetector_GPU::GoodFeaturesToTrackDetector_GPU(int maxCorners_, double qualityLevel_, double minDistance_,
int blockSize_, bool useHarrisDetector_, double harrisK_)
{
maxCorners = maxCorners_;
qualityLevel = qualityLevel_;
minDistance = minDistance_;
blockSize = blockSize_;
useHarrisDetector = useHarrisDetector_;
harrisK = harrisK_;
}
///////////////////////////// Mean Shift //////////////////////////////
//! Does mean shift filtering on GPU.
CV_EXPORTS void meanShiftFiltering(const GpuMat& src, GpuMat& dst, int sp, int sr,
TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1),
Stream& stream = Stream::Null());
//! Does mean shift procedure on GPU.
CV_EXPORTS void meanShiftProc(const GpuMat& src, GpuMat& dstr, GpuMat& dstsp, int sp, int sr,
TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1),
Stream& stream = Stream::Null());
//! Does mean shift segmentation with elimination of small regions.
CV_EXPORTS void meanShiftSegmentation(const GpuMat& src, Mat& dst, int sp, int sr, int minsize,
TermCriteria criteria = TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 5, 1));
/////////////////////////// Match Template ////////////////////////////
struct CV_EXPORTS MatchTemplateBuf
{
Size user_block_size;
GpuMat imagef, templf;
std::vector<GpuMat> images;
std::vector<GpuMat> image_sums;
std::vector<GpuMat> image_sqsums;
};
//! computes the proximity map for the raster template and the image where the template is searched for
CV_EXPORTS void matchTemplate(const GpuMat& image, const GpuMat& templ, GpuMat& result, int method, Stream &stream = Stream::Null());
//! computes the proximity map for the raster template and the image where the template is searched for
CV_EXPORTS void matchTemplate(const GpuMat& image, const GpuMat& templ, GpuMat& result, int method, MatchTemplateBuf &buf, Stream& stream = Stream::Null());
////////////////////////// Bilateral Filter ///////////////////////////
//! Performa bilateral filtering of passsed image
CV_EXPORTS void bilateralFilter(const GpuMat& src, GpuMat& dst, int kernel_size, float sigma_color, float sigma_spatial,
int borderMode = BORDER_DEFAULT, Stream& stream = Stream::Null());
///////////////////////////// Blending ////////////////////////////////
//! performs linear blending of two images
//! to avoid accuracy errors sum of weigths shouldn't be very close to zero
CV_EXPORTS void blendLinear(const GpuMat& img1, const GpuMat& img2, const GpuMat& weights1, const GpuMat& weights2,
GpuMat& result, Stream& stream = Stream::Null());
}} // namespace cv { namespace gpu {
#endif /* __OPENCV_GPUIMGPROC_HPP__ */

93
modules/gpuimgproc/perf/perf_bilateral_filter.cpp
View File

@@ -1,93 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// BilateralFilter
DEF_PARAM_TEST(Sz_Depth_Cn_KernelSz, cv::Size, MatDepth, MatCn, int);
PERF_TEST_P(Sz_Depth_Cn_KernelSz, BilateralFilter,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_32F),
GPU_CHANNELS_1_3,
Values(3, 5, 9)))
{
declare.time(60.0);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int kernel_size = GET_PARAM(3);
const float sigma_color = 7;
const float sigma_spatial = 5;
const int borderMode = cv::BORDER_REFLECT101;
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::bilateralFilter(d_src, dst, kernel_size, sigma_color, sigma_spatial, borderMode);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::bilateralFilter(src, dst, kernel_size, sigma_color, sigma_spatial, borderMode);
CPU_SANITY_CHECK(dst);
}
}

86
modules/gpuimgproc/perf/perf_blend.cpp
View File

@@ -1,86 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// BlendLinear
PERF_TEST_P(Sz_Depth_Cn, BlendLinear,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_32F),
GPU_CHANNELS_1_3_4))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat img1(size, type);
cv::Mat img2(size, type);
declare.in(img1, img2, WARMUP_RNG);
const cv::Mat weights1(size, CV_32FC1, cv::Scalar::all(0.5));
const cv::Mat weights2(size, CV_32FC1, cv::Scalar::all(0.5));
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_img1(img1);
const cv::gpu::GpuMat d_img2(img2);
const cv::gpu::GpuMat d_weights1(weights1);
const cv::gpu::GpuMat d_weights2(weights2);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::blendLinear(d_img1, d_img2, d_weights1, d_weights2, dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL_NO_CPU();
}
}

87
modules/gpuimgproc/perf/perf_canny.cpp
View File

@@ -1,87 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// Canny
DEF_PARAM_TEST(Image_AppertureSz_L2gradient, string, int, bool);
PERF_TEST_P(Image_AppertureSz_L2gradient, Canny,
Combine(Values("perf/800x600.png", "perf/1280x1024.png", "perf/1680x1050.png"),
Values(3, 5),
Bool()))
{
const string fileName = GET_PARAM(0);
const int apperture_size = GET_PARAM(1);
const bool useL2gradient = GET_PARAM(2);
const cv::Mat image = readImage(fileName, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
const double low_thresh = 50.0;
const double high_thresh = 100.0;
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_image(image);
cv::gpu::GpuMat dst;
cv::gpu::CannyBuf d_buf;
TEST_CYCLE() cv::gpu::Canny(d_image, d_buf, dst, low_thresh, high_thresh, apperture_size, useL2gradient);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::Canny(image, dst, low_thresh, high_thresh, apperture_size, useL2gradient);
CPU_SANITY_CHECK(dst);
}
}

252
modules/gpuimgproc/perf/perf_color.cpp
View File

@@ -1,252 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// CvtColor
DEF_PARAM_TEST(Sz_Depth_Code, cv::Size, MatDepth, CvtColorInfo);
PERF_TEST_P(Sz_Depth_Code, CvtColor,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_32F),
Values(CvtColorInfo(4, 4, cv::COLOR_RGBA2BGRA),
CvtColorInfo(4, 1, cv::COLOR_BGRA2GRAY),
CvtColorInfo(1, 4, cv::COLOR_GRAY2BGRA),
CvtColorInfo(3, 3, cv::COLOR_BGR2XYZ),
CvtColorInfo(3, 3, cv::COLOR_XYZ2BGR),
CvtColorInfo(3, 3, cv::COLOR_BGR2YCrCb),
CvtColorInfo(3, 3, cv::COLOR_YCrCb2BGR),
CvtColorInfo(3, 3, cv::COLOR_BGR2YUV),
CvtColorInfo(3, 3, cv::COLOR_YUV2BGR),
CvtColorInfo(3, 3, cv::COLOR_BGR2HSV),
CvtColorInfo(3, 3, cv::COLOR_HSV2BGR),
CvtColorInfo(3, 3, cv::COLOR_BGR2HLS),
CvtColorInfo(3, 3, cv::COLOR_HLS2BGR),
CvtColorInfo(3, 3, cv::COLOR_BGR2Lab),
CvtColorInfo(3, 3, cv::COLOR_LBGR2Lab),
CvtColorInfo(3, 3, cv::COLOR_BGR2Luv),
CvtColorInfo(3, 3, cv::COLOR_LBGR2Luv),
CvtColorInfo(3, 3, cv::COLOR_Lab2BGR),
CvtColorInfo(3, 3, cv::COLOR_Lab2LBGR),
CvtColorInfo(3, 3, cv::COLOR_Luv2RGB),
CvtColorInfo(3, 3, cv::COLOR_Luv2LRGB))))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const CvtColorInfo info = GET_PARAM(2);
cv::Mat src(size, CV_MAKETYPE(depth, info.scn));
cv::randu(src, 0, depth == CV_8U ? 255.0 : 1.0);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::cvtColor(d_src, dst, info.code, info.dcn);
GPU_SANITY_CHECK(dst, 1e-4);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::cvtColor(src, dst, info.code, info.dcn);
CPU_SANITY_CHECK(dst);
}
}
PERF_TEST_P(Sz_Depth_Code, CvtColorBayer,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U),
Values(CvtColorInfo(1, 3, cv::COLOR_BayerBG2BGR),
CvtColorInfo(1, 3, cv::COLOR_BayerGB2BGR),
CvtColorInfo(1, 3, cv::COLOR_BayerRG2BGR),
CvtColorInfo(1, 3, cv::COLOR_BayerGR2BGR),
CvtColorInfo(1, 1, cv::COLOR_BayerBG2GRAY),
CvtColorInfo(1, 1, cv::COLOR_BayerGB2GRAY),
CvtColorInfo(1, 1, cv::COLOR_BayerRG2GRAY),
CvtColorInfo(1, 1, cv::COLOR_BayerGR2GRAY))))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const CvtColorInfo info = GET_PARAM(2);
cv::Mat src(size, CV_MAKETYPE(depth, info.scn));
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::cvtColor(d_src, dst, info.code, info.dcn);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::cvtColor(src, dst, info.code, info.dcn);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// Demosaicing
CV_ENUM(DemosaicingCode,
cv::COLOR_BayerBG2BGR, cv::COLOR_BayerGB2BGR, cv::COLOR_BayerRG2BGR, cv::COLOR_BayerGR2BGR,
cv::COLOR_BayerBG2GRAY, cv::COLOR_BayerGB2GRAY, cv::COLOR_BayerRG2GRAY, cv::COLOR_BayerGR2GRAY,
cv::gpu::COLOR_BayerBG2BGR_MHT, cv::gpu::COLOR_BayerGB2BGR_MHT, cv::gpu::COLOR_BayerRG2BGR_MHT, cv::gpu::COLOR_BayerGR2BGR_MHT,
cv::gpu::COLOR_BayerBG2GRAY_MHT, cv::gpu::COLOR_BayerGB2GRAY_MHT, cv::gpu::COLOR_BayerRG2GRAY_MHT, cv::gpu::COLOR_BayerGR2GRAY_MHT)
DEF_PARAM_TEST(Sz_Code, cv::Size, DemosaicingCode);
PERF_TEST_P(Sz_Code, Demosaicing,
Combine(GPU_TYPICAL_MAT_SIZES,
DemosaicingCode::all()))
{
const cv::Size size = GET_PARAM(0);
const int code = GET_PARAM(1);
cv::Mat src(size, CV_8UC1);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::demosaicing(d_src, dst, code);
GPU_SANITY_CHECK(dst);
}
else
{
if (code >= cv::COLOR_COLORCVT_MAX)
{
FAIL_NO_CPU();
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::cvtColor(src, dst, code);
CPU_SANITY_CHECK(dst);
}
}
}
//////////////////////////////////////////////////////////////////////
// SwapChannels
PERF_TEST_P(Sz, SwapChannels,
GPU_TYPICAL_MAT_SIZES)
{
const cv::Size size = GetParam();
cv::Mat src(size, CV_8UC4);
declare.in(src, WARMUP_RNG);
const int dstOrder[] = {2, 1, 0, 3};
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat dst(src);
TEST_CYCLE() cv::gpu::swapChannels(dst, dstOrder);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// AlphaComp
CV_ENUM(AlphaOp, cv::gpu::ALPHA_OVER, cv::gpu::ALPHA_IN, cv::gpu::ALPHA_OUT, cv::gpu::ALPHA_ATOP, cv::gpu::ALPHA_XOR, cv::gpu::ALPHA_PLUS, cv::gpu::ALPHA_OVER_PREMUL, cv::gpu::ALPHA_IN_PREMUL, cv::gpu::ALPHA_OUT_PREMUL, cv::gpu::ALPHA_ATOP_PREMUL, cv::gpu::ALPHA_XOR_PREMUL, cv::gpu::ALPHA_PLUS_PREMUL, cv::gpu::ALPHA_PREMUL)
DEF_PARAM_TEST(Sz_Type_Op, cv::Size, MatType, AlphaOp);
PERF_TEST_P(Sz_Type_Op, AlphaComp,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8UC4, CV_16UC4, CV_32SC4, CV_32FC4),
AlphaOp::all()))
{
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int alpha_op = GET_PARAM(2);
cv::Mat img1(size, type);
cv::Mat img2(size, type);
declare.in(img1, img2, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_img1(img1);
const cv::gpu::GpuMat d_img2(img2);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::alphaComp(d_img1, d_img2, dst, alpha_op);
GPU_SANITY_CHECK(dst, 1e-3, ERROR_RELATIVE);
}
else
{
FAIL_NO_CPU();
}
}

137
modules/gpuimgproc/perf/perf_corners.cpp
View File

@@ -1,137 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// CornerHarris
DEF_PARAM_TEST(Image_Type_Border_BlockSz_ApertureSz, string, MatType, BorderMode, int, int);
PERF_TEST_P(Image_Type_Border_BlockSz_ApertureSz, CornerHarris,
Combine(Values<string>("gpu/stereobm/aloe-L.png"),
Values(CV_8UC1, CV_32FC1),
Values(BorderMode(cv::BORDER_REFLECT101), BorderMode(cv::BORDER_REPLICATE), BorderMode(cv::BORDER_REFLECT)),
Values(3, 5, 7),
Values(0, 3, 5, 7)))
{
const string fileName = GET_PARAM(0);
const int type = GET_PARAM(1);
const int borderMode = GET_PARAM(2);
const int blockSize = GET_PARAM(3);
const int apertureSize = GET_PARAM(4);
cv::Mat img = readImage(fileName, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
img.convertTo(img, type, type == CV_32F ? 1.0 / 255.0 : 1.0);
const double k = 0.5;
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_Dx;
cv::gpu::GpuMat d_Dy;
cv::gpu::GpuMat d_buf;
TEST_CYCLE() cv::gpu::cornerHarris(d_img, dst, d_Dx, d_Dy, d_buf, blockSize, apertureSize, k, borderMode);
GPU_SANITY_CHECK(dst, 1e-4);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::cornerHarris(img, dst, blockSize, apertureSize, k, borderMode);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// CornerMinEigenVal
PERF_TEST_P(Image_Type_Border_BlockSz_ApertureSz, CornerMinEigenVal,
Combine(Values<string>("gpu/stereobm/aloe-L.png"),
Values(CV_8UC1, CV_32FC1),
Values(BorderMode(cv::BORDER_REFLECT101), BorderMode(cv::BORDER_REPLICATE), BorderMode(cv::BORDER_REFLECT)),
Values(3, 5, 7),
Values(0, 3, 5, 7)))
{
const string fileName = GET_PARAM(0);
const int type = GET_PARAM(1);
const int borderMode = GET_PARAM(2);
const int blockSize = GET_PARAM(3);
const int apertureSize = GET_PARAM(4);
cv::Mat img = readImage(fileName, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
img.convertTo(img, type, type == CV_32F ? 1.0 / 255.0 : 1.0);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_Dx;
cv::gpu::GpuMat d_Dy;
cv::gpu::GpuMat d_buf;
TEST_CYCLE() cv::gpu::cornerMinEigenVal(d_img, dst, d_Dx, d_Dy, d_buf, blockSize, apertureSize, borderMode);
GPU_SANITY_CHECK(dst, 1e-4);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::cornerMinEigenVal(img, dst, blockSize, apertureSize, borderMode);
CPU_SANITY_CHECK(dst);
}
}

86
modules/gpuimgproc/perf/perf_gftt.cpp
View File

@@ -1,86 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////
// GoodFeaturesToTrack
DEF_PARAM_TEST(Image_MinDistance, string, double);
PERF_TEST_P(Image_MinDistance, GoodFeaturesToTrack,
Combine(Values<string>("gpu/perf/aloe.png"),
Values(0.0, 3.0)))
{
const string fileName = GET_PARAM(0);
const double minDistance = GET_PARAM(1);
const cv::Mat image = readImage(fileName, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
const int maxCorners = 8000;
const double qualityLevel = 0.01;
if (PERF_RUN_GPU())
{
cv::gpu::GoodFeaturesToTrackDetector_GPU d_detector(maxCorners, qualityLevel, minDistance);
const cv::gpu::GpuMat d_image(image);
cv::gpu::GpuMat pts;
TEST_CYCLE() d_detector(d_image, pts);
GPU_SANITY_CHECK(pts);
}
else
{
cv::Mat pts;
TEST_CYCLE() cv::goodFeaturesToTrack(image, pts, maxCorners, qualityLevel, minDistance);
CPU_SANITY_CHECK(pts);
}
}

221
modules/gpuimgproc/perf/perf_histogram.cpp
View File

@@ -1,221 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// HistEvenC1
PERF_TEST_P(Sz_Depth, HistEvenC1,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_16S)))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
cv::Mat src(size, depth);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf;
TEST_CYCLE() cv::gpu::histEven(d_src, dst, d_buf, 30, 0, 180);
GPU_SANITY_CHECK(dst);
}
else
{
const int hbins = 30;
const float hranges[] = {0.0f, 180.0f};
const int histSize[] = {hbins};
const float* ranges[] = {hranges};
const int channels[] = {0};
cv::Mat dst;
TEST_CYCLE() cv::calcHist(&src, 1, channels, cv::Mat(), dst, 1, histSize, ranges);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// HistEvenC4
PERF_TEST_P(Sz_Depth, HistEvenC4,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_16S)))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
cv::Mat src(size, CV_MAKE_TYPE(depth, 4));
declare.in(src, WARMUP_RNG);
int histSize[] = {30, 30, 30, 30};
int lowerLevel[] = {0, 0, 0, 0};
int upperLevel[] = {180, 180, 180, 180};
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_hist[4];
cv::gpu::GpuMat d_buf;
TEST_CYCLE() cv::gpu::histEven(d_src, d_hist, d_buf, histSize, lowerLevel, upperLevel);
cv::Mat cpu_hist0, cpu_hist1, cpu_hist2, cpu_hist3;
d_hist[0].download(cpu_hist0);
d_hist[1].download(cpu_hist1);
d_hist[2].download(cpu_hist2);
d_hist[3].download(cpu_hist3);
SANITY_CHECK(cpu_hist0);
SANITY_CHECK(cpu_hist1);
SANITY_CHECK(cpu_hist2);
SANITY_CHECK(cpu_hist3);
}
else
{
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// CalcHist
PERF_TEST_P(Sz, CalcHist,
GPU_TYPICAL_MAT_SIZES)
{
const cv::Size size = GetParam();
cv::Mat src(size, CV_8UC1);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::calcHist(d_src, dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// EqualizeHist
PERF_TEST_P(Sz, EqualizeHist,
GPU_TYPICAL_MAT_SIZES)
{
const cv::Size size = GetParam();
cv::Mat src(size, CV_8UC1);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_hist;
cv::gpu::GpuMat d_buf;
TEST_CYCLE() cv::gpu::equalizeHist(d_src, dst, d_hist, d_buf);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::equalizeHist(src, dst);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// CLAHE
DEF_PARAM_TEST(Sz_ClipLimit, cv::Size, double);
PERF_TEST_P(Sz_ClipLimit, CLAHE,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(0.0, 40.0)))
{
const cv::Size size = GET_PARAM(0);
const double clipLimit = GET_PARAM(1);
cv::Mat src(size, CV_8UC1);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::Ptr<cv::gpu::CLAHE> clahe = cv::gpu::createCLAHE(clipLimit);
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE() clahe->apply(d_src, dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Ptr<cv::CLAHE> clahe = cv::createCLAHE(clipLimit);
cv::Mat dst;
TEST_CYCLE() clahe->apply(src, dst);
CPU_SANITY_CHECK(dst);
}
}

317
modules/gpuimgproc/perf/perf_hough.cpp
View File

@@ -1,317 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// HoughLines
namespace
{
struct Vec4iComparator
{
bool operator()(const cv::Vec4i& a, const cv::Vec4i b) const
{
if (a[0] != b[0]) return a[0] < b[0];
else if(a[1] != b[1]) return a[1] < b[1];
else if(a[2] != b[2]) return a[2] < b[2];
else return a[3] < b[3];
}
};
struct Vec3fComparator
{
bool operator()(const cv::Vec3f& a, const cv::Vec3f b) const
{
if(a[0] != b[0]) return a[0] < b[0];
else if(a[1] != b[1]) return a[1] < b[1];
else return a[2] < b[2];
}
};
struct Vec2fComparator
{
bool operator()(const cv::Vec2f& a, const cv::Vec2f b) const
{
if(a[0] != b[0]) return a[0] < b[0];
else return a[1] < b[1];
}
};
}
PERF_TEST_P(Sz, HoughLines,
GPU_TYPICAL_MAT_SIZES)
{
declare.time(30.0);
const cv::Size size = GetParam();
const float rho = 1.0f;
const float theta = static_cast<float>(CV_PI / 180.0);
const int threshold = 300;
cv::Mat src(size, CV_8UC1, cv::Scalar::all(0));
cv::line(src, cv::Point(0, 100), cv::Point(src.cols, 100), cv::Scalar::all(255), 1);
cv::line(src, cv::Point(0, 200), cv::Point(src.cols, 200), cv::Scalar::all(255), 1);
cv::line(src, cv::Point(0, 400), cv::Point(src.cols, 400), cv::Scalar::all(255), 1);
cv::line(src, cv::Point(100, 0), cv::Point(100, src.rows), cv::Scalar::all(255), 1);
cv::line(src, cv::Point(200, 0), cv::Point(200, src.rows), cv::Scalar::all(255), 1);
cv::line(src, cv::Point(400, 0), cv::Point(400, src.rows), cv::Scalar::all(255), 1);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_lines;
cv::gpu::HoughLinesBuf d_buf;
TEST_CYCLE() cv::gpu::HoughLines(d_src, d_lines, d_buf, rho, theta, threshold);
cv::Mat gpu_lines(d_lines.row(0));
cv::Vec2f* begin = gpu_lines.ptr<cv::Vec2f>(0);
cv::Vec2f* end = begin + gpu_lines.cols;
std::sort(begin, end, Vec2fComparator());
SANITY_CHECK(gpu_lines);
}
else
{
std::vector<cv::Vec2f> cpu_lines;
TEST_CYCLE() cv::HoughLines(src, cpu_lines, rho, theta, threshold);
SANITY_CHECK(cpu_lines);
}
}
//////////////////////////////////////////////////////////////////////
// HoughLinesP
DEF_PARAM_TEST_1(Image, std::string);
PERF_TEST_P(Image, HoughLinesP,
testing::Values("cv/shared/pic5.png", "stitching/a1.png"))
{
declare.time(30.0);
const std::string fileName = getDataPath(GetParam());
const float rho = 1.0f;
const float theta = static_cast<float>(CV_PI / 180.0);
const int threshold = 100;
const int minLineLenght = 50;
const int maxLineGap = 5;
const cv::Mat image = cv::imread(fileName, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
cv::Mat mask;
cv::Canny(image, mask, 50, 100);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_mask(mask);
cv::gpu::GpuMat d_lines;
cv::gpu::HoughLinesBuf d_buf;
TEST_CYCLE() cv::gpu::HoughLinesP(d_mask, d_lines, d_buf, rho, theta, minLineLenght, maxLineGap);
cv::Mat gpu_lines(d_lines);
cv::Vec4i* begin = gpu_lines.ptr<cv::Vec4i>();
cv::Vec4i* end = begin + gpu_lines.cols;
std::sort(begin, end, Vec4iComparator());
SANITY_CHECK(gpu_lines);
}
else
{
std::vector<cv::Vec4i> cpu_lines;
TEST_CYCLE() cv::HoughLinesP(mask, cpu_lines, rho, theta, threshold, minLineLenght, maxLineGap);
SANITY_CHECK(cpu_lines);
}
}
//////////////////////////////////////////////////////////////////////
// HoughCircles
DEF_PARAM_TEST(Sz_Dp_MinDist, cv::Size, float, float);
PERF_TEST_P(Sz_Dp_MinDist, HoughCircles,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(1.0f, 2.0f, 4.0f),
Values(1.0f)))
{
declare.time(30.0);
const cv::Size size = GET_PARAM(0);
const float dp = GET_PARAM(1);
const float minDist = GET_PARAM(2);
const int minRadius = 10;
const int maxRadius = 30;
const int cannyThreshold = 100;
const int votesThreshold = 15;
cv::Mat src(size, CV_8UC1, cv::Scalar::all(0));
cv::circle(src, cv::Point(100, 100), 20, cv::Scalar::all(255), -1);
cv::circle(src, cv::Point(200, 200), 25, cv::Scalar::all(255), -1);
cv::circle(src, cv::Point(200, 100), 25, cv::Scalar::all(255), -1);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_circles;
cv::gpu::HoughCirclesBuf d_buf;
TEST_CYCLE() cv::gpu::HoughCircles(d_src, d_circles, d_buf, cv::HOUGH_GRADIENT, dp, minDist, cannyThreshold, votesThreshold, minRadius, maxRadius);
cv::Mat gpu_circles(d_circles);
cv::Vec3f* begin = gpu_circles.ptr<cv::Vec3f>(0);
cv::Vec3f* end = begin + gpu_circles.cols;
std::sort(begin, end, Vec3fComparator());
SANITY_CHECK(gpu_circles);
}
else
{
std::vector<cv::Vec3f> cpu_circles;
TEST_CYCLE() cv::HoughCircles(src, cpu_circles, cv::HOUGH_GRADIENT, dp, minDist, cannyThreshold, votesThreshold, minRadius, maxRadius);
SANITY_CHECK(cpu_circles);
}
}
//////////////////////////////////////////////////////////////////////
// GeneralizedHough
enum { GHT_POSITION = cv::GeneralizedHough::GHT_POSITION,
GHT_SCALE = cv::GeneralizedHough::GHT_SCALE,
GHT_ROTATION = cv::GeneralizedHough::GHT_ROTATION
};
CV_FLAGS(GHMethod, GHT_POSITION, GHT_SCALE, GHT_ROTATION);
DEF_PARAM_TEST(Method_Sz, GHMethod, cv::Size);
PERF_TEST_P(Method_Sz, GeneralizedHough,
Combine(Values(GHMethod(GHT_POSITION), GHMethod(GHT_POSITION | GHT_SCALE), GHMethod(GHT_POSITION | GHT_ROTATION), GHMethod(GHT_POSITION | GHT_SCALE | GHT_ROTATION)),
GPU_TYPICAL_MAT_SIZES))
{
declare.time(10);
const int method = GET_PARAM(0);
const cv::Size imageSize = GET_PARAM(1);
const cv::Mat templ = readImage("cv/shared/templ.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(templ.empty());
cv::Mat image(imageSize, CV_8UC1, cv::Scalar::all(0));
templ.copyTo(image(cv::Rect(50, 50, templ.cols, templ.rows)));
cv::RNG rng(123456789);
const int objCount = rng.uniform(5, 15);
for (int i = 0; i < objCount; ++i)
{
double scale = rng.uniform(0.7, 1.3);
bool rotate = 1 == rng.uniform(0, 2);
cv::Mat obj;
cv::resize(templ, obj, cv::Size(), scale, scale);
if (rotate)
obj = obj.t();
cv::Point pos;
pos.x = rng.uniform(0, image.cols - obj.cols);
pos.y = rng.uniform(0, image.rows - obj.rows);
cv::Mat roi = image(cv::Rect(pos, obj.size()));
cv::add(roi, obj, roi);
}
cv::Mat edges;
cv::Canny(image, edges, 50, 100);
cv::Mat dx, dy;
cv::Sobel(image, dx, CV_32F, 1, 0);
cv::Sobel(image, dy, CV_32F, 0, 1);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_edges(edges);
const cv::gpu::GpuMat d_dx(dx);
const cv::gpu::GpuMat d_dy(dy);
cv::gpu::GpuMat posAndVotes;
cv::Ptr<cv::gpu::GeneralizedHough_GPU> d_hough = cv::gpu::GeneralizedHough_GPU::create(method);
if (method & GHT_ROTATION)
{
d_hough->set("maxAngle", 90.0);
d_hough->set("angleStep", 2.0);
}
d_hough->setTemplate(cv::gpu::GpuMat(templ));
TEST_CYCLE() d_hough->detect(d_edges, d_dx, d_dy, posAndVotes);
const cv::gpu::GpuMat positions(1, posAndVotes.cols, CV_32FC4, posAndVotes.data);
GPU_SANITY_CHECK(positions);
}
else
{
cv::Mat positions;
cv::Ptr<cv::GeneralizedHough> hough = cv::GeneralizedHough::create(method);
if (method & GHT_ROTATION)
{
hough->set("maxAngle", 90.0);
hough->set("angleStep", 2.0);
}
hough->setTemplate(templ);
TEST_CYCLE() hough->detect(edges, dx, dy, positions);
CPU_SANITY_CHECK(positions);
}
}

47
modules/gpuimgproc/perf/perf_main.cpp
View File

@@ -1,47 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace perf;
CV_PERF_TEST_MAIN(gpuimgproc, printCudaInfo())

131
modules/gpuimgproc/perf/perf_match_template.cpp
View File

@@ -1,131 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate8U
CV_ENUM(TemplateMethod, TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED)
DEF_PARAM_TEST(Sz_TemplateSz_Cn_Method, cv::Size, cv::Size, MatCn, TemplateMethod);
PERF_TEST_P(Sz_TemplateSz_Cn_Method, MatchTemplate8U,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(cv::Size(5, 5), cv::Size(16, 16), cv::Size(30, 30)),
GPU_CHANNELS_1_3_4,
TemplateMethod::all()))
{
declare.time(300.0);
const cv::Size size = GET_PARAM(0);
const cv::Size templ_size = GET_PARAM(1);
const int cn = GET_PARAM(2);
const int method = GET_PARAM(3);
cv::Mat image(size, CV_MAKE_TYPE(CV_8U, cn));
cv::Mat templ(templ_size, CV_MAKE_TYPE(CV_8U, cn));
declare.in(image, templ, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_image(image);
const cv::gpu::GpuMat d_templ(templ);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::matchTemplate(d_image, d_templ, dst, method);
GPU_SANITY_CHECK(dst, 1e-5, ERROR_RELATIVE);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::matchTemplate(image, templ, dst, method);
CPU_SANITY_CHECK(dst);
}
};
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate32F
PERF_TEST_P(Sz_TemplateSz_Cn_Method, MatchTemplate32F,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(cv::Size(5, 5), cv::Size(16, 16), cv::Size(30, 30)),
GPU_CHANNELS_1_3_4,
Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_CCORR))))
{
declare.time(300.0);
const cv::Size size = GET_PARAM(0);
const cv::Size templ_size = GET_PARAM(1);
const int cn = GET_PARAM(2);
int method = GET_PARAM(3);
cv::Mat image(size, CV_MAKE_TYPE(CV_32F, cn));
cv::Mat templ(templ_size, CV_MAKE_TYPE(CV_32F, cn));
declare.in(image, templ, WARMUP_RNG);
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_image(image);
const cv::gpu::GpuMat d_templ(templ);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::matchTemplate(d_image, d_templ, dst, method);
GPU_SANITY_CHECK(dst, 1e-6, ERROR_RELATIVE);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::matchTemplate(image, templ, dst, method);
CPU_SANITY_CHECK(dst);
}
}

152
modules/gpuimgproc/perf/perf_mean_shift.cpp
View File

@@ -1,152 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace std;
using namespace testing;
using namespace perf;
//////////////////////////////////////////////////////////////////////
// MeanShiftFiltering
DEF_PARAM_TEST_1(Image, string);
PERF_TEST_P(Image, MeanShiftFiltering,
Values<string>("gpu/meanshift/cones.png"))
{
declare.time(300.0);
const cv::Mat img = readImage(GetParam());
ASSERT_FALSE(img.empty());
cv::Mat rgba;
cv::cvtColor(img, rgba, cv::COLOR_BGR2BGRA);
const int sp = 50;
const int sr = 50;
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(rgba);
cv::gpu::GpuMat dst;
TEST_CYCLE() cv::gpu::meanShiftFiltering(d_src, dst, sp, sr);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::pyrMeanShiftFiltering(img, dst, sp, sr);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// MeanShiftProc
PERF_TEST_P(Image, MeanShiftProc,
Values<string>("gpu/meanshift/cones.png"))
{
declare.time(300.0);
const cv::Mat img = readImage(GetParam());
ASSERT_FALSE(img.empty());
cv::Mat rgba;
cv::cvtColor(img, rgba, cv::COLOR_BGR2BGRA);
const int sp = 50;
const int sr = 50;
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(rgba);
cv::gpu::GpuMat dstr;
cv::gpu::GpuMat dstsp;
TEST_CYCLE() cv::gpu::meanShiftProc(d_src, dstr, dstsp, sp, sr);
GPU_SANITY_CHECK(dstr);
GPU_SANITY_CHECK(dstsp);
}
else
{
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// MeanShiftSegmentation
PERF_TEST_P(Image, MeanShiftSegmentation,
Values<string>("gpu/meanshift/cones.png"))
{
declare.time(300.0);
const cv::Mat img = readImage(GetParam());
ASSERT_FALSE(img.empty());
cv::Mat rgba;
cv::cvtColor(img, rgba, cv::COLOR_BGR2BGRA);
const int sp = 10;
const int sr = 10;
const int minsize = 20;
if (PERF_RUN_GPU())
{
const cv::gpu::GpuMat d_src(rgba);
cv::Mat dst;
TEST_CYCLE() cv::gpu::meanShiftSegmentation(d_src, dst, sp, sr, minsize);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL_NO_CPU();
}
}

43
modules/gpuimgproc/perf/perf_precomp.cpp
View File

@@ -1,43 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"

64
modules/gpuimgproc/perf/perf_precomp.hpp
View File

@@ -1,64 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifdef __GNUC__
# pragma GCC diagnostic ignored "-Wmissing-declarations"
# if defined __clang__ || defined __APPLE__
# pragma GCC diagnostic ignored "-Wmissing-prototypes"
# pragma GCC diagnostic ignored "-Wextra"
# endif
#endif
#ifndef __OPENCV_PERF_PRECOMP_HPP__
#define __OPENCV_PERF_PRECOMP_HPP__
#include "opencv2/ts.hpp"
#include "opencv2/ts/gpu_perf.hpp"
#include "opencv2/gpuimgproc.hpp"
#include "opencv2/imgproc.hpp"
#ifdef GTEST_CREATE_SHARED_LIBRARY
#error no modules except ts should have GTEST_CREATE_SHARED_LIBRARY defined
#endif
#endif

96
modules/gpuimgproc/src/bilateral_filter.cpp
View File

@@ -1,96 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::bilateralFilter(const GpuMat&, GpuMat&, int, float, float, int, Stream&) { throw_no_cuda(); }
#else
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
template<typename T>
void bilateral_filter_gpu(const PtrStepSzb& src, PtrStepSzb dst, int kernel_size, float sigma_spatial, float sigma_color, int borderMode, cudaStream_t stream);
}
}}}
void cv::gpu::bilateralFilter(const GpuMat& src, GpuMat& dst, int kernel_size, float sigma_color, float sigma_spatial, int borderMode, Stream& s)
{
using cv::gpu::cudev::imgproc::bilateral_filter_gpu;
typedef void (*func_t)(const PtrStepSzb& src, PtrStepSzb dst, int kernel_size, float sigma_spatial, float sigma_color, int borderMode, cudaStream_t s);
static const func_t funcs[6][4] =
{
{bilateral_filter_gpu<uchar> , 0 /*bilateral_filter_gpu<uchar2>*/ , bilateral_filter_gpu<uchar3> , bilateral_filter_gpu<uchar4> },
{0 /*bilateral_filter_gpu<schar>*/, 0 /*bilateral_filter_gpu<schar2>*/ , 0 /*bilateral_filter_gpu<schar3>*/, 0 /*bilateral_filter_gpu<schar4>*/},
{bilateral_filter_gpu<ushort> , 0 /*bilateral_filter_gpu<ushort2>*/, bilateral_filter_gpu<ushort3> , bilateral_filter_gpu<ushort4> },
{bilateral_filter_gpu<short> , 0 /*bilateral_filter_gpu<short2>*/ , bilateral_filter_gpu<short3> , bilateral_filter_gpu<short4> },
{0 /*bilateral_filter_gpu<int>*/ , 0 /*bilateral_filter_gpu<int2>*/ , 0 /*bilateral_filter_gpu<int3>*/ , 0 /*bilateral_filter_gpu<int4>*/ },
{bilateral_filter_gpu<float> , 0 /*bilateral_filter_gpu<float2>*/ , bilateral_filter_gpu<float3> , bilateral_filter_gpu<float4> }
};
sigma_color = (sigma_color <= 0 ) ? 1 : sigma_color;
sigma_spatial = (sigma_spatial <= 0 ) ? 1 : sigma_spatial;
int radius = (kernel_size <= 0) ? cvRound(sigma_spatial*1.5) : kernel_size/2;
kernel_size = std::max(radius, 1)*2 + 1;
CV_Assert(src.depth() <= CV_32F && src.channels() <= 4);
const func_t func = funcs[src.depth()][src.channels() - 1];
CV_Assert(func != 0);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
dst.create(src.size(), src.type());
func(src, dst, kernel_size, sigma_spatial, sigma_color, borderMode, StreamAccessor::getStream(s));
}
#endif

102
modules/gpuimgproc/src/blend.cpp
View File

@@ -1,102 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::blendLinear(const GpuMat&, const GpuMat&, const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
#else
////////////////////////////////////////////////////////////////////////
// blendLinear
namespace cv { namespace gpu { namespace cudev
{
namespace blend
{
template <typename T>
void blendLinearCaller(int rows, int cols, int cn, PtrStep<T> img1, PtrStep<T> img2, PtrStepf weights1, PtrStepf weights2, PtrStep<T> result, cudaStream_t stream);
void blendLinearCaller8UC4(int rows, int cols, PtrStepb img1, PtrStepb img2, PtrStepf weights1, PtrStepf weights2, PtrStepb result, cudaStream_t stream);
}
}}}
using namespace ::cv::gpu::cudev::blend;
void cv::gpu::blendLinear(const GpuMat& img1, const GpuMat& img2, const GpuMat& weights1, const GpuMat& weights2,
GpuMat& result, Stream& stream)
{
CV_Assert(img1.size() == img2.size());
CV_Assert(img1.type() == img2.type());
CV_Assert(weights1.size() == img1.size());
CV_Assert(weights2.size() == img2.size());
CV_Assert(weights1.type() == CV_32F);
CV_Assert(weights2.type() == CV_32F);
const Size size = img1.size();
const int depth = img1.depth();
const int cn = img1.channels();
result.create(size, CV_MAKE_TYPE(depth, cn));
switch (depth)
{
case CV_8U:
if (cn != 4)
blendLinearCaller<uchar>(size.height, size.width, cn, img1, img2, weights1, weights2, result, StreamAccessor::getStream(stream));
else
blendLinearCaller8UC4(size.height, size.width, img1, img2, weights1, weights2, result, StreamAccessor::getStream(stream));
break;
case CV_32F:
blendLinearCaller<float>(size.height, size.width, cn, img1, img2, weights1, weights2, result, StreamAccessor::getStream(stream));
break;
default:
CV_Error(cv::Error::StsUnsupportedFormat, "bad image depth in linear blending function");
}
}
#endif

186
modules/gpuimgproc/src/canny.cpp
View File

@@ -1,186 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::Canny(const GpuMat&, GpuMat&, double, double, int, bool) { throw_no_cuda(); }
void cv::gpu::Canny(const GpuMat&, CannyBuf&, GpuMat&, double, double, int, bool) { throw_no_cuda(); }
void cv::gpu::Canny(const GpuMat&, const GpuMat&, GpuMat&, double, double, bool) { throw_no_cuda(); }
void cv::gpu::Canny(const GpuMat&, const GpuMat&, CannyBuf&, GpuMat&, double, double, bool) { throw_no_cuda(); }
void cv::gpu::CannyBuf::create(const Size&, int) { throw_no_cuda(); }
void cv::gpu::CannyBuf::release() { throw_no_cuda(); }
#else /* !defined (HAVE_CUDA) */
void cv::gpu::CannyBuf::create(const Size& image_size, int apperture_size)
{
if (apperture_size > 0)
{
ensureSizeIsEnough(image_size, CV_32SC1, dx);
ensureSizeIsEnough(image_size, CV_32SC1, dy);
if (apperture_size != 3)
{
filterDX = createDerivFilter_GPU(CV_8UC1, CV_32S, 1, 0, apperture_size, BORDER_REPLICATE);
filterDY = createDerivFilter_GPU(CV_8UC1, CV_32S, 0, 1, apperture_size, BORDER_REPLICATE);
}
}
ensureSizeIsEnough(image_size, CV_32FC1, mag);
ensureSizeIsEnough(image_size, CV_32SC1, map);
ensureSizeIsEnough(1, image_size.area(), CV_16UC2, st1);
ensureSizeIsEnough(1, image_size.area(), CV_16UC2, st2);
}
void cv::gpu::CannyBuf::release()
{
dx.release();
dy.release();
mag.release();
map.release();
st1.release();
st2.release();
}
namespace canny
{
void calcMagnitude(PtrStepSzb srcWhole, int xoff, int yoff, PtrStepSzi dx, PtrStepSzi dy, PtrStepSzf mag, bool L2Grad);
void calcMagnitude(PtrStepSzi dx, PtrStepSzi dy, PtrStepSzf mag, bool L2Grad);
void calcMap(PtrStepSzi dx, PtrStepSzi dy, PtrStepSzf mag, PtrStepSzi map, float low_thresh, float high_thresh);
void edgesHysteresisLocal(PtrStepSzi map, ushort2* st1);
void edgesHysteresisGlobal(PtrStepSzi map, ushort2* st1, ushort2* st2);
void getEdges(PtrStepSzi map, PtrStepSzb dst);
}
namespace
{
void CannyCaller(const GpuMat& dx, const GpuMat& dy, CannyBuf& buf, GpuMat& dst, float low_thresh, float high_thresh)
{
using namespace canny;
buf.map.setTo(Scalar::all(0));
calcMap(dx, dy, buf.mag, buf.map, low_thresh, high_thresh);
edgesHysteresisLocal(buf.map, buf.st1.ptr<ushort2>());
edgesHysteresisGlobal(buf.map, buf.st1.ptr<ushort2>(), buf.st2.ptr<ushort2>());
getEdges(buf.map, dst);
}
}
void cv::gpu::Canny(const GpuMat& src, GpuMat& dst, double low_thresh, double high_thresh, int apperture_size, bool L2gradient)
{
CannyBuf buf;
Canny(src, buf, dst, low_thresh, high_thresh, apperture_size, L2gradient);
}
void cv::gpu::Canny(const GpuMat& src, CannyBuf& buf, GpuMat& dst, double low_thresh, double high_thresh, int apperture_size, bool L2gradient)
{
using namespace canny;
CV_Assert(src.type() == CV_8UC1);
if (!deviceSupports(SHARED_ATOMICS))
CV_Error(cv::Error::StsNotImplemented, "The device doesn't support shared atomics");
if( low_thresh > high_thresh )
std::swap( low_thresh, high_thresh);
dst.create(src.size(), CV_8U);
buf.create(src.size(), apperture_size);
if (apperture_size == 3)
{
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
GpuMat srcWhole(wholeSize, src.type(), src.datastart, src.step);
calcMagnitude(srcWhole, ofs.x, ofs.y, buf.dx, buf.dy, buf.mag, L2gradient);
}
else
{
buf.filterDX->apply(src, buf.dx, Rect(0, 0, src.cols, src.rows));
buf.filterDY->apply(src, buf.dy, Rect(0, 0, src.cols, src.rows));
calcMagnitude(buf.dx, buf.dy, buf.mag, L2gradient);
}
CannyCaller(buf.dx, buf.dy, buf, dst, static_cast<float>(low_thresh), static_cast<float>(high_thresh));
}
void cv::gpu::Canny(const GpuMat& dx, const GpuMat& dy, GpuMat& dst, double low_thresh, double high_thresh, bool L2gradient)
{
CannyBuf buf;
Canny(dx, dy, buf, dst, low_thresh, high_thresh, L2gradient);
}
void cv::gpu::Canny(const GpuMat& dx, const GpuMat& dy, CannyBuf& buf, GpuMat& dst, double low_thresh, double high_thresh, bool L2gradient)
{
using namespace canny;
CV_Assert(TargetArchs::builtWith(SHARED_ATOMICS) && DeviceInfo().supports(SHARED_ATOMICS));
CV_Assert(dx.type() == CV_32SC1 && dy.type() == CV_32SC1 && dx.size() == dy.size());
if( low_thresh > high_thresh )
std::swap( low_thresh, high_thresh);
dst.create(dx.size(), CV_8U);
buf.create(dx.size(), -1);
calcMagnitude(dx, dy, buf.mag, L2gradient);
CannyCaller(dx, dy, buf, dst, static_cast<float>(low_thresh), static_cast<float>(high_thresh));
}
#endif /* !defined (HAVE_CUDA) */

2080
modules/gpuimgproc/src/color.cpp
View File

File diff suppressed because it is too large Load Diff

149
modules/gpuimgproc/src/corners.cpp
View File

@@ -1,149 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::cornerHarris(const GpuMat&, GpuMat&, int, int, double, int) { throw_no_cuda(); }
void cv::gpu::cornerHarris(const GpuMat&, GpuMat&, GpuMat&, GpuMat&, int, int, double, int) { throw_no_cuda(); }
void cv::gpu::cornerHarris(const GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, int, int, double, int, Stream&) { throw_no_cuda(); }
void cv::gpu::cornerMinEigenVal(const GpuMat&, GpuMat&, int, int, int) { throw_no_cuda(); }
void cv::gpu::cornerMinEigenVal(const GpuMat&, GpuMat&, GpuMat&, GpuMat&, int, int, int) { throw_no_cuda(); }
void cv::gpu::cornerMinEigenVal(const GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, int, int, int, Stream&) { throw_no_cuda(); }
#else /* !defined (HAVE_CUDA) */
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
void cornerHarris_gpu(int block_size, float k, PtrStepSzf Dx, PtrStepSzf Dy, PtrStepSzf dst, int border_type, cudaStream_t stream);
void cornerMinEigenVal_gpu(int block_size, PtrStepSzf Dx, PtrStepSzf Dy, PtrStepSzf dst, int border_type, cudaStream_t stream);
}
}}}
namespace
{
void extractCovData(const GpuMat& src, GpuMat& Dx, GpuMat& Dy, GpuMat& buf, int blockSize, int ksize, int borderType, Stream& stream)
{
double scale = static_cast<double>(1 << ((ksize > 0 ? ksize : 3) - 1)) * blockSize;
if (ksize < 0)
scale *= 2.;
if (src.depth() == CV_8U)
scale *= 255.;
scale = 1./scale;
Dx.create(src.size(), CV_32F);
Dy.create(src.size(), CV_32F);
if (ksize > 0)
{
Sobel(src, Dx, CV_32F, 1, 0, buf, ksize, scale, borderType, -1, stream);
Sobel(src, Dy, CV_32F, 0, 1, buf, ksize, scale, borderType, -1, stream);
}
else
{
Scharr(src, Dx, CV_32F, 1, 0, buf, scale, borderType, -1, stream);
Scharr(src, Dy, CV_32F, 0, 1, buf, scale, borderType, -1, stream);
}
}
}
void cv::gpu::cornerHarris(const GpuMat& src, GpuMat& dst, int blockSize, int ksize, double k, int borderType)
{
GpuMat Dx, Dy;
cornerHarris(src, dst, Dx, Dy, blockSize, ksize, k, borderType);
}
void cv::gpu::cornerHarris(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, int blockSize, int ksize, double k, int borderType)
{
GpuMat buf;
cornerHarris(src, dst, Dx, Dy, buf, blockSize, ksize, k, borderType);
}
void cv::gpu::cornerHarris(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, GpuMat& buf, int blockSize, int ksize, double k, int borderType, Stream& stream)
{
using namespace cv::gpu::cudev::imgproc;
CV_Assert(borderType == cv::BORDER_REFLECT101 || borderType == cv::BORDER_REPLICATE || borderType == cv::BORDER_REFLECT);
extractCovData(src, Dx, Dy, buf, blockSize, ksize, borderType, stream);
dst.create(src.size(), CV_32F);
cornerHarris_gpu(blockSize, static_cast<float>(k), Dx, Dy, dst, borderType, StreamAccessor::getStream(stream));
}
void cv::gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, int blockSize, int ksize, int borderType)
{
GpuMat Dx, Dy;
cornerMinEigenVal(src, dst, Dx, Dy, blockSize, ksize, borderType);
}
void cv::gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, int blockSize, int ksize, int borderType)
{
GpuMat buf;
cornerMinEigenVal(src, dst, Dx, Dy, buf, blockSize, ksize, borderType);
}
void cv::gpu::cornerMinEigenVal(const GpuMat& src, GpuMat& dst, GpuMat& Dx, GpuMat& Dy, GpuMat& buf, int blockSize, int ksize, int borderType, Stream& stream)
{
using namespace ::cv::gpu::cudev::imgproc;
CV_Assert(borderType == cv::BORDER_REFLECT101 || borderType == cv::BORDER_REPLICATE || borderType == cv::BORDER_REFLECT);
extractCovData(src, Dx, Dy, buf, blockSize, ksize, borderType, stream);
dst.create(src.size(), CV_32F);
cornerMinEigenVal_gpu(blockSize, Dx, Dy, dst, borderType, StreamAccessor::getStream(stream));
}
#endif /* !defined (HAVE_CUDA) */

199
modules/gpuimgproc/src/cuda/bilateral_filter.cu
View File

@@ -1,199 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_traits.hpp"
#include "opencv2/core/cuda/vec_math.hpp"
#include "opencv2/core/cuda/border_interpolate.hpp"
using namespace cv::gpu;
typedef unsigned char uchar;
typedef unsigned short ushort;
//////////////////////////////////////////////////////////////////////////////////
/// Bilateral filtering
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
__device__ __forceinline__ float norm_l1(const float& a) { return ::fabs(a); }
__device__ __forceinline__ float norm_l1(const float2& a) { return ::fabs(a.x) + ::fabs(a.y); }
__device__ __forceinline__ float norm_l1(const float3& a) { return ::fabs(a.x) + ::fabs(a.y) + ::fabs(a.z); }
__device__ __forceinline__ float norm_l1(const float4& a) { return ::fabs(a.x) + ::fabs(a.y) + ::fabs(a.z) + ::fabs(a.w); }
__device__ __forceinline__ float sqr(const float& a) { return a * a; }
template<typename T, typename B>
__global__ void bilateral_kernel(const PtrStepSz<T> src, PtrStep<T> dst, const B b, const int ksz, const float sigma_spatial2_inv_half, const float sigma_color2_inv_half)
{
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type value_type;
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
if (x >= src.cols || y >= src.rows)
return;
value_type center = saturate_cast<value_type>(src(y, x));
value_type sum1 = VecTraits<value_type>::all(0);
float sum2 = 0;
int r = ksz / 2;
float r2 = (float)(r * r);
int tx = x - r + ksz;
int ty = y - r + ksz;
if (x - ksz/2 >=0 && y - ksz/2 >=0 && tx < src.cols && ty < src.rows)
{
for (int cy = y - r; cy < ty; ++cy)
for (int cx = x - r; cx < tx; ++cx)
{
float space2 = (x - cx) * (x - cx) + (y - cy) * (y - cy);
if (space2 > r2)
continue;
value_type value = saturate_cast<value_type>(src(cy, cx));
float weight = ::exp(space2 * sigma_spatial2_inv_half + sqr(norm_l1(value - center)) * sigma_color2_inv_half);
sum1 = sum1 + weight * value;
sum2 = sum2 + weight;
}
}
else
{
for (int cy = y - r; cy < ty; ++cy)
for (int cx = x - r; cx < tx; ++cx)
{
float space2 = (x - cx) * (x - cx) + (y - cy) * (y - cy);
if (space2 > r2)
continue;
value_type value = saturate_cast<value_type>(b.at(cy, cx, src.data, src.step));
float weight = ::exp(space2 * sigma_spatial2_inv_half + sqr(norm_l1(value - center)) * sigma_color2_inv_half);
sum1 = sum1 + weight * value;
sum2 = sum2 + weight;
}
}
dst(y, x) = saturate_cast<T>(sum1 / sum2);
}
template<typename T, template <typename> class B>
void bilateral_caller(const PtrStepSzb& src, PtrStepSzb dst, int kernel_size, float sigma_spatial, float sigma_color, cudaStream_t stream)
{
dim3 block (32, 8);
dim3 grid (divUp (src.cols, block.x), divUp (src.rows, block.y));
B<T> b(src.rows, src.cols);
float sigma_spatial2_inv_half = -0.5f/(sigma_spatial * sigma_spatial);
float sigma_color2_inv_half = -0.5f/(sigma_color * sigma_color);
cudaSafeCall( cudaFuncSetCacheConfig (bilateral_kernel<T, B<T> >, cudaFuncCachePreferL1) );
bilateral_kernel<<<grid, block>>>((PtrStepSz<T>)src, (PtrStepSz<T>)dst, b, kernel_size, sigma_spatial2_inv_half, sigma_color2_inv_half);
cudaSafeCall ( cudaGetLastError () );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template<typename T>
void bilateral_filter_gpu(const PtrStepSzb& src, PtrStepSzb dst, int kernel_size, float gauss_spatial_coeff, float gauss_color_coeff, int borderMode, cudaStream_t stream)
{
typedef void (*caller_t)(const PtrStepSzb& src, PtrStepSzb dst, int kernel_size, float sigma_spatial, float sigma_color, cudaStream_t stream);
static caller_t funcs[] =
{
bilateral_caller<T, BrdConstant>,
bilateral_caller<T, BrdReplicate>,
bilateral_caller<T, BrdReflect>,
bilateral_caller<T, BrdWrap>,
bilateral_caller<T, BrdReflect101>
};
funcs[borderMode](src, dst, kernel_size, gauss_spatial_coeff, gauss_color_coeff, stream);
}
}
}}}
#define OCV_INSTANTIATE_BILATERAL_FILTER(T) \
template void cv::gpu::cudev::imgproc::bilateral_filter_gpu<T>(const PtrStepSzb&, PtrStepSzb, int, float, float, int, cudaStream_t);
OCV_INSTANTIATE_BILATERAL_FILTER(uchar)
//OCV_INSTANTIATE_BILATERAL_FILTER(uchar2)
OCV_INSTANTIATE_BILATERAL_FILTER(uchar3)
OCV_INSTANTIATE_BILATERAL_FILTER(uchar4)
//OCV_INSTANTIATE_BILATERAL_FILTER(schar)
//OCV_INSTANTIATE_BILATERAL_FILTER(schar2)
//OCV_INSTANTIATE_BILATERAL_FILTER(schar3)
//OCV_INSTANTIATE_BILATERAL_FILTER(schar4)
OCV_INSTANTIATE_BILATERAL_FILTER(short)
//OCV_INSTANTIATE_BILATERAL_FILTER(short2)
OCV_INSTANTIATE_BILATERAL_FILTER(short3)
OCV_INSTANTIATE_BILATERAL_FILTER(short4)
OCV_INSTANTIATE_BILATERAL_FILTER(ushort)
//OCV_INSTANTIATE_BILATERAL_FILTER(ushort2)
OCV_INSTANTIATE_BILATERAL_FILTER(ushort3)
OCV_INSTANTIATE_BILATERAL_FILTER(ushort4)
//OCV_INSTANTIATE_BILATERAL_FILTER(int)
//OCV_INSTANTIATE_BILATERAL_FILTER(int2)
//OCV_INSTANTIATE_BILATERAL_FILTER(int3)
//OCV_INSTANTIATE_BILATERAL_FILTER(int4)
OCV_INSTANTIATE_BILATERAL_FILTER(float)
//OCV_INSTANTIATE_BILATERAL_FILTER(float2)
OCV_INSTANTIATE_BILATERAL_FILTER(float3)
OCV_INSTANTIATE_BILATERAL_FILTER(float4)
#endif /* CUDA_DISABLER */

121
modules/gpuimgproc/src/cuda/blend.cu
View File

@@ -1,121 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
namespace cv { namespace gpu { namespace cudev
{
namespace blend
{
template <typename T>
__global__ void blendLinearKernel(int rows, int cols, int cn, const PtrStep<T> img1, const PtrStep<T> img2,
const PtrStepf weights1, const PtrStepf weights2, PtrStep<T> result)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y < rows && x < cols)
{
int x_ = x / cn;
float w1 = weights1.ptr(y)[x_];
float w2 = weights2.ptr(y)[x_];
T p1 = img1.ptr(y)[x];
T p2 = img2.ptr(y)[x];
result.ptr(y)[x] = (p1 * w1 + p2 * w2) / (w1 + w2 + 1e-5f);
}
}
template <typename T>
void blendLinearCaller(int rows, int cols, int cn, PtrStep<T> img1, PtrStep<T> img2, PtrStepf weights1, PtrStepf weights2, PtrStep<T> result, cudaStream_t stream)
{
dim3 threads(16, 16);
dim3 grid(divUp(cols * cn, threads.x), divUp(rows, threads.y));
blendLinearKernel<<<grid, threads, 0, stream>>>(rows, cols * cn, cn, img1, img2, weights1, weights2, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall(cudaDeviceSynchronize());
}
template void blendLinearCaller<uchar>(int, int, int, PtrStep<uchar>, PtrStep<uchar>, PtrStepf, PtrStepf, PtrStep<uchar>, cudaStream_t stream);
template void blendLinearCaller<float>(int, int, int, PtrStep<float>, PtrStep<float>, PtrStepf, PtrStepf, PtrStep<float>, cudaStream_t stream);
__global__ void blendLinearKernel8UC4(int rows, int cols, const PtrStepb img1, const PtrStepb img2,
const PtrStepf weights1, const PtrStepf weights2, PtrStepb result)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y < rows && x < cols)
{
float w1 = weights1.ptr(y)[x];
float w2 = weights2.ptr(y)[x];
float sum_inv = 1.f / (w1 + w2 + 1e-5f);
w1 *= sum_inv;
w2 *= sum_inv;
uchar4 p1 = ((const uchar4*)img1.ptr(y))[x];
uchar4 p2 = ((const uchar4*)img2.ptr(y))[x];
((uchar4*)result.ptr(y))[x] = make_uchar4(p1.x * w1 + p2.x * w2, p1.y * w1 + p2.y * w2,
p1.z * w1 + p2.z * w2, p1.w * w1 + p2.w * w2);
}
}
void blendLinearCaller8UC4(int rows, int cols, PtrStepb img1, PtrStepb img2, PtrStepf weights1, PtrStepf weights2, PtrStepb result, cudaStream_t stream)
{
dim3 threads(16, 16);
dim3 grid(divUp(cols, threads.x), divUp(rows, threads.y));
blendLinearKernel8UC4<<<grid, threads, 0, stream>>>(rows, cols, img1, img2, weights1, weights2, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall(cudaDeviceSynchronize());
}
} // namespace blend
}}} // namespace cv { namespace gpu { namespace cudev
#endif /* CUDA_DISABLER */

494
modules/gpuimgproc/src/cuda/canny.cu
View File

@@ -1,494 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include <utility>
#include <algorithm>//std::swap
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/emulation.hpp"
#include "opencv2/core/cuda/transform.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "opencv2/core/cuda/utility.hpp"
using namespace cv::gpu;
using namespace cv::gpu::cudev;
namespace canny
{
struct L1 : binary_function<int, int, float>
{
__device__ __forceinline__ float operator ()(int x, int y) const
{
return ::abs(x) + ::abs(y);
}
__device__ __forceinline__ L1() {}
__device__ __forceinline__ L1(const L1&) {}
};
struct L2 : binary_function<int, int, float>
{
__device__ __forceinline__ float operator ()(int x, int y) const
{
return ::sqrtf(x * x + y * y);
}
__device__ __forceinline__ L2() {}
__device__ __forceinline__ L2(const L2&) {}
};
}
namespace cv { namespace gpu { namespace cudev
{
template <> struct TransformFunctorTraits<canny::L1> : DefaultTransformFunctorTraits<canny::L1>
{
enum { smart_shift = 4 };
};
template <> struct TransformFunctorTraits<canny::L2> : DefaultTransformFunctorTraits<canny::L2>
{
enum { smart_shift = 4 };
};
}}}
namespace canny
{
texture<uchar, cudaTextureType2D, cudaReadModeElementType> tex_src(false, cudaFilterModePoint, cudaAddressModeClamp);
struct SrcTex
{
int xoff;
int yoff;
__host__ SrcTex(int _xoff, int _yoff) : xoff(_xoff), yoff(_yoff) {}
__device__ __forceinline__ int operator ()(int y, int x) const
{
return tex2D(tex_src, x + xoff, y + yoff);
}
};
template <class Norm> __global__
void calcMagnitudeKernel(const SrcTex src, PtrStepi dx, PtrStepi dy, PtrStepSzf mag, const Norm norm)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y >= mag.rows || x >= mag.cols)
return;
int dxVal = (src(y - 1, x + 1) + 2 * src(y, x + 1) + src(y + 1, x + 1)) - (src(y - 1, x - 1) + 2 * src(y, x - 1) + src(y + 1, x - 1));
int dyVal = (src(y + 1, x - 1) + 2 * src(y + 1, x) + src(y + 1, x + 1)) - (src(y - 1, x - 1) + 2 * src(y - 1, x) + src(y - 1, x + 1));
dx(y, x) = dxVal;
dy(y, x) = dyVal;
mag(y, x) = norm(dxVal, dyVal);
}
void calcMagnitude(PtrStepSzb srcWhole, int xoff, int yoff, PtrStepSzi dx, PtrStepSzi dy, PtrStepSzf mag, bool L2Grad)
{
const dim3 block(16, 16);
const dim3 grid(divUp(mag.cols, block.x), divUp(mag.rows, block.y));
bindTexture(&tex_src, srcWhole);
SrcTex src(xoff, yoff);
if (L2Grad)
{
L2 norm;
calcMagnitudeKernel<<<grid, block>>>(src, dx, dy, mag, norm);
}
else
{
L1 norm;
calcMagnitudeKernel<<<grid, block>>>(src, dx, dy, mag, norm);
}
cudaSafeCall( cudaGetLastError() );
cudaSafeCall(cudaThreadSynchronize());
}
void calcMagnitude(PtrStepSzi dx, PtrStepSzi dy, PtrStepSzf mag, bool L2Grad)
{
if (L2Grad)
{
L2 norm;
transform(dx, dy, mag, norm, WithOutMask(), 0);
}
else
{
L1 norm;
transform(dx, dy, mag, norm, WithOutMask(), 0);
}
}
}
//////////////////////////////////////////////////////////////////////////////////////////
namespace canny
{
texture<float, cudaTextureType2D, cudaReadModeElementType> tex_mag(false, cudaFilterModePoint, cudaAddressModeClamp);
__global__ void calcMapKernel(const PtrStepSzi dx, const PtrStepi dy, PtrStepi map, const float low_thresh, const float high_thresh)
{
const int CANNY_SHIFT = 15;
const int TG22 = (int)(0.4142135623730950488016887242097*(1<<CANNY_SHIFT) + 0.5);
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x == 0 || x >= dx.cols - 1 || y == 0 || y >= dx.rows - 1)
return;
int dxVal = dx(y, x);
int dyVal = dy(y, x);
const int s = (dxVal ^ dyVal) < 0 ? -1 : 1;
const float m = tex2D(tex_mag, x, y);
dxVal = ::abs(dxVal);
dyVal = ::abs(dyVal);
// 0 - the pixel can not belong to an edge
// 1 - the pixel might belong to an edge
// 2 - the pixel does belong to an edge
int edge_type = 0;
if (m > low_thresh)
{
const int tg22x = dxVal * TG22;
const int tg67x = tg22x + ((dxVal + dxVal) << CANNY_SHIFT);
dyVal <<= CANNY_SHIFT;
if (dyVal < tg22x)
{
if (m > tex2D(tex_mag, x - 1, y) && m >= tex2D(tex_mag, x + 1, y))
edge_type = 1 + (int)(m > high_thresh);
}
else if(dyVal > tg67x)
{
if (m > tex2D(tex_mag, x, y - 1) && m >= tex2D(tex_mag, x, y + 1))
edge_type = 1 + (int)(m > high_thresh);
}
else
{
if (m > tex2D(tex_mag, x - s, y - 1) && m >= tex2D(tex_mag, x + s, y + 1))
edge_type = 1 + (int)(m > high_thresh);
}
}
map(y, x) = edge_type;
}
void calcMap(PtrStepSzi dx, PtrStepSzi dy, PtrStepSzf mag, PtrStepSzi map, float low_thresh, float high_thresh)
{
const dim3 block(16, 16);
const dim3 grid(divUp(dx.cols, block.x), divUp(dx.rows, block.y));
bindTexture(&tex_mag, mag);
calcMapKernel<<<grid, block>>>(dx, dy, map, low_thresh, high_thresh);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
}
//////////////////////////////////////////////////////////////////////////////////////////
namespace canny
{
__device__ int counter = 0;
__global__ void edgesHysteresisLocalKernel(PtrStepSzi map, ushort2* st)
{
__shared__ volatile int smem[18][18];
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
smem[threadIdx.y + 1][threadIdx.x + 1] = x < map.cols && y < map.rows ? map(y, x) : 0;
if (threadIdx.y == 0)
smem[0][threadIdx.x + 1] = y > 0 ? map(y - 1, x) : 0;
if (threadIdx.y == blockDim.y - 1)
smem[blockDim.y + 1][threadIdx.x + 1] = y + 1 < map.rows ? map(y + 1, x) : 0;
if (threadIdx.x == 0)
smem[threadIdx.y + 1][0] = x > 0 ? map(y, x - 1) : 0;
if (threadIdx.x == blockDim.x - 1)
smem[threadIdx.y + 1][blockDim.x + 1] = x + 1 < map.cols ? map(y, x + 1) : 0;
if (threadIdx.x == 0 && threadIdx.y == 0)
smem[0][0] = y > 0 && x > 0 ? map(y - 1, x - 1) : 0;
if (threadIdx.x == blockDim.x - 1 && threadIdx.y == 0)
smem[0][blockDim.x + 1] = y > 0 && x + 1 < map.cols ? map(y - 1, x + 1) : 0;
if (threadIdx.x == 0 && threadIdx.y == blockDim.y - 1)
smem[blockDim.y + 1][0] = y + 1 < map.rows && x > 0 ? map(y + 1, x - 1) : 0;
if (threadIdx.x == blockDim.x - 1 && threadIdx.y == blockDim.y - 1)
smem[blockDim.y + 1][blockDim.x + 1] = y + 1 < map.rows && x + 1 < map.cols ? map(y + 1, x + 1) : 0;
__syncthreads();
if (x >= map.cols || y >= map.rows)
return;
int n;
#pragma unroll
for (int k = 0; k < 16; ++k)
{
n = 0;
if (smem[threadIdx.y + 1][threadIdx.x + 1] == 1)
{
n += smem[threadIdx.y ][threadIdx.x ] == 2;
n += smem[threadIdx.y ][threadIdx.x + 1] == 2;
n += smem[threadIdx.y ][threadIdx.x + 2] == 2;
n += smem[threadIdx.y + 1][threadIdx.x ] == 2;
n += smem[threadIdx.y + 1][threadIdx.x + 2] == 2;
n += smem[threadIdx.y + 2][threadIdx.x ] == 2;
n += smem[threadIdx.y + 2][threadIdx.x + 1] == 2;
n += smem[threadIdx.y + 2][threadIdx.x + 2] == 2;
}
if (n > 0)
smem[threadIdx.y + 1][threadIdx.x + 1] = 2;
}
const int e = smem[threadIdx.y + 1][threadIdx.x + 1];
map(y, x) = e;
n = 0;
if (e == 2)
{
n += smem[threadIdx.y ][threadIdx.x ] == 1;
n += smem[threadIdx.y ][threadIdx.x + 1] == 1;
n += smem[threadIdx.y ][threadIdx.x + 2] == 1;
n += smem[threadIdx.y + 1][threadIdx.x ] == 1;
n += smem[threadIdx.y + 1][threadIdx.x + 2] == 1;
n += smem[threadIdx.y + 2][threadIdx.x ] == 1;
n += smem[threadIdx.y + 2][threadIdx.x + 1] == 1;
n += smem[threadIdx.y + 2][threadIdx.x + 2] == 1;
}
if (n > 0)
{
const int ind = ::atomicAdd(&counter, 1);
st[ind] = make_ushort2(x, y);
}
}
void edgesHysteresisLocal(PtrStepSzi map, ushort2* st1)
{
void* counter_ptr;
cudaSafeCall( cudaGetSymbolAddress(&counter_ptr, counter) );
cudaSafeCall( cudaMemset(counter_ptr, 0, sizeof(int)) );
const dim3 block(16, 16);
const dim3 grid(divUp(map.cols, block.x), divUp(map.rows, block.y));
edgesHysteresisLocalKernel<<<grid, block>>>(map, st1);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
}
//////////////////////////////////////////////////////////////////////////////////////////
namespace canny
{
__constant__ int c_dx[8] = {-1, 0, 1, -1, 1, -1, 0, 1};
__constant__ int c_dy[8] = {-1, -1, -1, 0, 0, 1, 1, 1};
__global__ void edgesHysteresisGlobalKernel(PtrStepSzi map, ushort2* st1, ushort2* st2, const int count)
{
const int stack_size = 512;
__shared__ int s_counter;
__shared__ int s_ind;
__shared__ ushort2 s_st[stack_size];
if (threadIdx.x == 0)
s_counter = 0;
__syncthreads();
int ind = blockIdx.y * gridDim.x + blockIdx.x;
if (ind >= count)
return;
ushort2 pos = st1[ind];
if (threadIdx.x < 8)
{
pos.x += c_dx[threadIdx.x];
pos.y += c_dy[threadIdx.x];
if (pos.x > 0 && pos.x < map.cols && pos.y > 0 && pos.y < map.rows && map(pos.y, pos.x) == 1)
{
map(pos.y, pos.x) = 2;
ind = Emulation::smem::atomicAdd(&s_counter, 1);
s_st[ind] = pos;
}
}
__syncthreads();
while (s_counter > 0 && s_counter <= stack_size - blockDim.x)
{
const int subTaskIdx = threadIdx.x >> 3;
const int portion = ::min(s_counter, blockDim.x >> 3);
if (subTaskIdx < portion)
pos = s_st[s_counter - 1 - subTaskIdx];
__syncthreads();
if (threadIdx.x == 0)
s_counter -= portion;
__syncthreads();
if (subTaskIdx < portion)
{
pos.x += c_dx[threadIdx.x & 7];
pos.y += c_dy[threadIdx.x & 7];
if (pos.x > 0 && pos.x < map.cols && pos.y > 0 && pos.y < map.rows && map(pos.y, pos.x) == 1)
{
map(pos.y, pos.x) = 2;
ind = Emulation::smem::atomicAdd(&s_counter, 1);
s_st[ind] = pos;
}
}
__syncthreads();
}
if (s_counter > 0)
{
if (threadIdx.x == 0)
{
ind = ::atomicAdd(&counter, s_counter);
s_ind = ind - s_counter;
}
__syncthreads();
ind = s_ind;
for (int i = threadIdx.x; i < s_counter; i += blockDim.x)
st2[ind + i] = s_st[i];
}
}
void edgesHysteresisGlobal(PtrStepSzi map, ushort2* st1, ushort2* st2)
{
void* counter_ptr;
cudaSafeCall( cudaGetSymbolAddress(&counter_ptr, canny::counter) );
int count;
cudaSafeCall( cudaMemcpy(&count, counter_ptr, sizeof(int), cudaMemcpyDeviceToHost) );
while (count > 0)
{
cudaSafeCall( cudaMemset(counter_ptr, 0, sizeof(int)) );
const dim3 block(128);
const dim3 grid(::min(count, 65535u), divUp(count, 65535), 1);
edgesHysteresisGlobalKernel<<<grid, block>>>(map, st1, st2, count);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaMemcpy(&count, counter_ptr, sizeof(int), cudaMemcpyDeviceToHost) );
std::swap(st1, st2);
}
}
}
//////////////////////////////////////////////////////////////////////////////////////////
namespace canny
{
struct GetEdges : unary_function<int, uchar>
{
__device__ __forceinline__ uchar operator ()(int e) const
{
return (uchar)(-(e >> 1));
}
__device__ __forceinline__ GetEdges() {}
__device__ __forceinline__ GetEdges(const GetEdges&) {}
};
}
namespace cv { namespace gpu { namespace cudev
{
template <> struct TransformFunctorTraits<canny::GetEdges> : DefaultTransformFunctorTraits<canny::GetEdges>
{
enum { smart_shift = 4 };
};
}}}
namespace canny
{
void getEdges(PtrStepSzi map, PtrStepSzb dst)
{
transform(map, dst, GetEdges(), WithOutMask(), 0);
}
}
#endif /* CUDA_DISABLER */

186
modules/gpuimgproc/src/cuda/clahe.cu
View File

@@ -1,186 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "opencv2/core/cuda/emulation.hpp"
#include "opencv2/core/cuda/scan.hpp"
#include "opencv2/core/cuda/reduce.hpp"
#include "opencv2/core/cuda/saturate_cast.hpp"
using namespace cv::gpu;
using namespace cv::gpu::cudev;
namespace clahe
{
__global__ void calcLutKernel(const PtrStepb src, PtrStepb lut,
const int2 tileSize, const int tilesX,
const int clipLimit, const float lutScale)
{
__shared__ int smem[512];
const int tx = blockIdx.x;
const int ty = blockIdx.y;
const unsigned int tid = threadIdx.y * blockDim.x + threadIdx.x;
smem[tid] = 0;
__syncthreads();
for (int i = threadIdx.y; i < tileSize.y; i += blockDim.y)
{
const uchar* srcPtr = src.ptr(ty * tileSize.y + i) + tx * tileSize.x;
for (int j = threadIdx.x; j < tileSize.x; j += blockDim.x)
{
const int data = srcPtr[j];
Emulation::smem::atomicAdd(&smem[data], 1);
}
}
__syncthreads();
int tHistVal = smem[tid];
__syncthreads();
if (clipLimit > 0)
{
// clip histogram bar
int clipped = 0;
if (tHistVal > clipLimit)
{
clipped = tHistVal - clipLimit;
tHistVal = clipLimit;
}
// find number of overall clipped samples
reduce<256>(smem, clipped, tid, plus<int>());
// broadcast evaluated value
__shared__ int totalClipped;
if (tid == 0)
totalClipped = clipped;
__syncthreads();
// redistribute clipped samples evenly
int redistBatch = totalClipped / 256;
tHistVal += redistBatch;
int residual = totalClipped - redistBatch * 256;
if (tid < residual)
++tHistVal;
}
const int lutVal = blockScanInclusive<256>(tHistVal, smem, tid);
lut(ty * tilesX + tx, tid) = saturate_cast<uchar>(__float2int_rn(lutScale * lutVal));
}
void calcLut(PtrStepSzb src, PtrStepb lut, int tilesX, int tilesY, int2 tileSize, int clipLimit, float lutScale, cudaStream_t stream)
{
const dim3 block(32, 8);
const dim3 grid(tilesX, tilesY);
calcLutKernel<<<grid, block, 0, stream>>>(src, lut, tileSize, tilesX, clipLimit, lutScale);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void tranformKernel(const PtrStepSzb src, PtrStepb dst, const PtrStepb lut, const int2 tileSize, const int tilesX, const int tilesY)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x >= src.cols || y >= src.rows)
return;
const float tyf = (static_cast<float>(y) / tileSize.y) - 0.5f;
int ty1 = __float2int_rd(tyf);
int ty2 = ty1 + 1;
const float ya = tyf - ty1;
ty1 = ::max(ty1, 0);
ty2 = ::min(ty2, tilesY - 1);
const float txf = (static_cast<float>(x) / tileSize.x) - 0.5f;
int tx1 = __float2int_rd(txf);
int tx2 = tx1 + 1;
const float xa = txf - tx1;
tx1 = ::max(tx1, 0);
tx2 = ::min(tx2, tilesX - 1);
const int srcVal = src(y, x);
float res = 0;
res += lut(ty1 * tilesX + tx1, srcVal) * ((1.0f - xa) * (1.0f - ya));
res += lut(ty1 * tilesX + tx2, srcVal) * ((xa) * (1.0f - ya));
res += lut(ty2 * tilesX + tx1, srcVal) * ((1.0f - xa) * (ya));
res += lut(ty2 * tilesX + tx2, srcVal) * ((xa) * (ya));
dst(y, x) = saturate_cast<uchar>(res);
}
void transform(PtrStepSzb src, PtrStepSzb dst, PtrStepb lut, int tilesX, int tilesY, int2 tileSize, cudaStream_t stream)
{
const dim3 block(32, 8);
const dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(tranformKernel, cudaFuncCachePreferL1) );
tranformKernel<<<grid, block, 0, stream>>>(src, dst, lut, tileSize, tilesX, tilesY);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
#endif // CUDA_DISABLER

461
modules/gpuimgproc/src/cuda/color.cu
View File

@@ -1,461 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/transform.hpp"
#include "opencv2/core/cuda/color.hpp"
#include "cvt_color_internal.h"
namespace cv { namespace gpu { namespace cudev
{
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_rgba_traits<uchar>::functor_type)
{
enum { smart_block_dim_x = 8 };
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_bgr555_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_bgr555_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_bgr565_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_bgr565_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgr555_to_bgra_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgr555_to_rgba_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgr565_to_bgra_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgr565_to_rgba_traits::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(gray_to_bgra_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(gray_to_bgr555_traits::functor_type)
{
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(gray_to_bgr565_traits::functor_type)
{
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_yuv4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_yuv4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(yuv4_to_bgra_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(yuv4_to_rgba_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_YCrCb4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_YCrCb4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(YCrCb4_to_bgra_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(YCrCb4_to_rgba_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_xyz4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_xyz4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(xyz4_to_bgra_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(xyz4_to_rgba_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_hsv4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_hsv4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(hsv4_to_bgra_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(hsv4_to_rgba_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(bgra_to_hls4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(rgba_to_hls4_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(hls4_to_bgra_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
OPENCV_GPU_TRANSFORM_FUNCTOR_TRAITS(hls4_to_rgba_traits<uchar>::functor_type)
{
enum { smart_block_dim_y = 8 };
enum { smart_shift = 4 };
};
#define OPENCV_GPU_IMPLEMENT_CVTCOLOR(name, traits) \
void name(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream) \
{ \
traits::functor_type functor = traits::create_functor(); \
typedef typename traits::functor_type::argument_type src_t; \
typedef typename traits::functor_type::result_type dst_t; \
cv::gpu::cudev::transform((PtrStepSz<src_t>)src, (PtrStepSz<dst_t>)dst, functor, WithOutMask(), stream); \
}
#define OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(name) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name, name ## _traits)
#define OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(name) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _8u, name ## _traits<uchar>) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _16u, name ## _traits<ushort>) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _32f, name ## _traits<float>)
#define OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(name) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _8u, name ## _traits<uchar>) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _32f, name ## _traits<float>)
#define OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(name) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _8u, name ## _traits<uchar>) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _32f, name ## _traits<float>) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _full_8u, name ## _full_traits<uchar>) \
OPENCV_GPU_IMPLEMENT_CVTCOLOR(name ## _full_32f, name ## _full_traits<float>)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr_to_bgr555)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr_to_bgr565)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(rgb_to_bgr555)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(rgb_to_bgr565)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgra_to_bgr555)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgra_to_bgr565)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(rgba_to_bgr555)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(rgba_to_bgr565)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr555_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr565_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr555_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr565_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr555_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr565_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr555_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr565_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(gray_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(gray_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(gray_to_bgr555)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(gray_to_bgr565)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr555_to_gray)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE(bgr565_to_gray)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_gray)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_gray)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_gray)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_gray)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_yuv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_yuv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_yuv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_yuv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_yuv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_yuv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_yuv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_yuv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(yuv4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_YCrCb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_YCrCb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_YCrCb4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_YCrCb4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_YCrCb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_YCrCb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_YCrCb4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_YCrCb4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(YCrCb4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_xyz)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_xyz)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgb_to_xyz4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(rgba_to_xyz4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_xyz)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_xyz)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgr_to_xyz4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(bgra_to_xyz4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL(xyz4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgb_to_hsv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgba_to_hsv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgb_to_hsv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgba_to_hsv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgr_to_hsv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgra_to_hsv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgr_to_hsv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgra_to_hsv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hsv4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgb_to_hls)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgba_to_hls)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgb_to_hls4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(rgba_to_hls4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgr_to_hls)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgra_to_hls)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgr_to_hls4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(bgra_to_hls4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL(hls4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgb_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgba_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgb_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgba_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgr_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgra_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgr_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgra_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgb_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgba_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgb_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgba_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgr_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgra_to_lab)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgr_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgra_to_lab4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_lrgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_lrgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_lrgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_lrgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_lbgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_lbgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab_to_lbgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lab4_to_lbgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgb_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgba_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgb_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(rgba_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgr_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgra_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgr_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(bgra_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgb_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgba_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgb_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lrgba_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgr_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgra_to_luv)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgr_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(lbgra_to_luv4)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_rgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_rgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_bgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_bgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_lrgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_lrgb)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_lrgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_lrgba)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_lbgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_lbgr)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv_to_lbgra)
OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F(luv4_to_lbgra)
#undef OPENCV_GPU_IMPLEMENT_CVTCOLOR
#undef OPENCV_GPU_IMPLEMENT_CVTCOLOR_ONE
#undef OPENCV_GPU_IMPLEMENT_CVTCOLOR_ALL
#undef OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F
#undef OPENCV_GPU_IMPLEMENT_CVTCOLOR_8U32F_FULL
}}} // namespace cv { namespace gpu { namespace cudev
#endif /* CUDA_DISABLER */

274
modules/gpuimgproc/src/cuda/corners.cu
View File

@@ -1,274 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_traits.hpp"
#include "opencv2/core/cuda/vec_math.hpp"
#include "opencv2/core/cuda/saturate_cast.hpp"
#include "opencv2/core/cuda/border_interpolate.hpp"
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
/////////////////////////////////////////// Corner Harris /////////////////////////////////////////////////
texture<float, cudaTextureType2D, cudaReadModeElementType> harrisDxTex(0, cudaFilterModePoint, cudaAddressModeClamp);
texture<float, cudaTextureType2D, cudaReadModeElementType> harrisDyTex(0, cudaFilterModePoint, cudaAddressModeClamp);
__global__ void cornerHarris_kernel(const int block_size, const float k, PtrStepSzf dst)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < dst.cols && y < dst.rows)
{
float a = 0.f;
float b = 0.f;
float c = 0.f;
const int ibegin = y - (block_size / 2);
const int jbegin = x - (block_size / 2);
const int iend = ibegin + block_size;
const int jend = jbegin + block_size;
for (int i = ibegin; i < iend; ++i)
{
for (int j = jbegin; j < jend; ++j)
{
float dx = tex2D(harrisDxTex, j, i);
float dy = tex2D(harrisDyTex, j, i);
a += dx * dx;
b += dx * dy;
c += dy * dy;
}
}
dst(y, x) = a * c - b * b - k * (a + c) * (a + c);
}
}
template <typename BR, typename BC>
__global__ void cornerHarris_kernel(const int block_size, const float k, PtrStepSzf dst, const BR border_row, const BC border_col)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < dst.cols && y < dst.rows)
{
float a = 0.f;
float b = 0.f;
float c = 0.f;
const int ibegin = y - (block_size / 2);
const int jbegin = x - (block_size / 2);
const int iend = ibegin + block_size;
const int jend = jbegin + block_size;
for (int i = ibegin; i < iend; ++i)
{
const int y = border_col.idx_row(i);
for (int j = jbegin; j < jend; ++j)
{
const int x = border_row.idx_col(j);
float dx = tex2D(harrisDxTex, x, y);
float dy = tex2D(harrisDyTex, x, y);
a += dx * dx;
b += dx * dy;
c += dy * dy;
}
}
dst(y, x) = a * c - b * b - k * (a + c) * (a + c);
}
}
void cornerHarris_gpu(int block_size, float k, PtrStepSzf Dx, PtrStepSzf Dy, PtrStepSzf dst, int border_type, cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(Dx.cols, block.x), divUp(Dx.rows, block.y));
bindTexture(&harrisDxTex, Dx);
bindTexture(&harrisDyTex, Dy);
switch (border_type)
{
case BORDER_REFLECT101:
cornerHarris_kernel<<<grid, block, 0, stream>>>(block_size, k, dst, BrdRowReflect101<void>(Dx.cols), BrdColReflect101<void>(Dx.rows));
break;
case BORDER_REFLECT:
cornerHarris_kernel<<<grid, block, 0, stream>>>(block_size, k, dst, BrdRowReflect<void>(Dx.cols), BrdColReflect<void>(Dx.rows));
break;
case BORDER_REPLICATE:
cornerHarris_kernel<<<grid, block, 0, stream>>>(block_size, k, dst);
break;
}
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
/////////////////////////////////////////// Corner Min Eigen Val /////////////////////////////////////////////////
texture<float, cudaTextureType2D, cudaReadModeElementType> minEigenValDxTex(0, cudaFilterModePoint, cudaAddressModeClamp);
texture<float, cudaTextureType2D, cudaReadModeElementType> minEigenValDyTex(0, cudaFilterModePoint, cudaAddressModeClamp);
__global__ void cornerMinEigenVal_kernel(const int block_size, PtrStepSzf dst)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < dst.cols && y < dst.rows)
{
float a = 0.f;
float b = 0.f;
float c = 0.f;
const int ibegin = y - (block_size / 2);
const int jbegin = x - (block_size / 2);
const int iend = ibegin + block_size;
const int jend = jbegin + block_size;
for (int i = ibegin; i < iend; ++i)
{
for (int j = jbegin; j < jend; ++j)
{
float dx = tex2D(minEigenValDxTex, j, i);
float dy = tex2D(minEigenValDyTex, j, i);
a += dx * dx;
b += dx * dy;
c += dy * dy;
}
}
a *= 0.5f;
c *= 0.5f;
dst(y, x) = (a + c) - sqrtf((a - c) * (a - c) + b * b);
}
}
template <typename BR, typename BC>
__global__ void cornerMinEigenVal_kernel(const int block_size, PtrStepSzf dst, const BR border_row, const BC border_col)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < dst.cols && y < dst.rows)
{
float a = 0.f;
float b = 0.f;
float c = 0.f;
const int ibegin = y - (block_size / 2);
const int jbegin = x - (block_size / 2);
const int iend = ibegin + block_size;
const int jend = jbegin + block_size;
for (int i = ibegin; i < iend; ++i)
{
int y = border_col.idx_row(i);
for (int j = jbegin; j < jend; ++j)
{
int x = border_row.idx_col(j);
float dx = tex2D(minEigenValDxTex, x, y);
float dy = tex2D(minEigenValDyTex, x, y);
a += dx * dx;
b += dx * dy;
c += dy * dy;
}
}
a *= 0.5f;
c *= 0.5f;
dst(y, x) = (a + c) - sqrtf((a - c) * (a - c) + b * b);
}
}
void cornerMinEigenVal_gpu(int block_size, PtrStepSzf Dx, PtrStepSzf Dy, PtrStepSzf dst, int border_type, cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(Dx.cols, block.x), divUp(Dx.rows, block.y));
bindTexture(&minEigenValDxTex, Dx);
bindTexture(&minEigenValDyTex, Dy);
switch (border_type)
{
case BORDER_REFLECT101:
cornerMinEigenVal_kernel<<<grid, block, 0, stream>>>(block_size, dst, BrdRowReflect101<void>(Dx.cols), BrdColReflect101<void>(Dx.rows));
break;
case BORDER_REFLECT:
cornerMinEigenVal_kernel<<<grid, block, 0, stream>>>(block_size, dst, BrdRowReflect<void>(Dx.cols), BrdColReflect<void>(Dx.rows));
break;
case BORDER_REPLICATE:
cornerMinEigenVal_kernel<<<grid, block, 0, stream>>>(block_size, dst);
break;
}
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall(cudaDeviceSynchronize());
}
}
}}}
#endif

544
modules/gpuimgproc/src/cuda/debayer.cu
View File

@@ -1,544 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_traits.hpp"
#include "opencv2/core/cuda/vec_math.hpp"
#include "opencv2/core/cuda/limits.hpp"
#include "opencv2/core/cuda/color.hpp"
#include "opencv2/core/cuda/saturate_cast.hpp"
namespace cv { namespace gpu { namespace cudev
{
template <typename T> struct Bayer2BGR;
template <> struct Bayer2BGR<uchar>
{
uchar3 res0;
uchar3 res1;
uchar3 res2;
uchar3 res3;
__device__ void apply(const PtrStepSzb& src, int s_x, int s_y, bool blue_last, bool start_with_green)
{
uchar4 patch[3][3];
patch[0][1] = ((const uchar4*) src.ptr(s_y - 1))[s_x];
patch[0][0] = ((const uchar4*) src.ptr(s_y - 1))[::max(s_x - 1, 0)];
patch[0][2] = ((const uchar4*) src.ptr(s_y - 1))[::min(s_x + 1, ((src.cols + 3) >> 2) - 1)];
patch[1][1] = ((const uchar4*) src.ptr(s_y))[s_x];
patch[1][0] = ((const uchar4*) src.ptr(s_y))[::max(s_x - 1, 0)];
patch[1][2] = ((const uchar4*) src.ptr(s_y))[::min(s_x + 1, ((src.cols + 3) >> 2) - 1)];
patch[2][1] = ((const uchar4*) src.ptr(s_y + 1))[s_x];
patch[2][0] = ((const uchar4*) src.ptr(s_y + 1))[::max(s_x - 1, 0)];
patch[2][2] = ((const uchar4*) src.ptr(s_y + 1))[::min(s_x + 1, ((src.cols + 3) >> 2) - 1)];
if ((s_y & 1) ^ start_with_green)
{
const int t0 = (patch[0][1].x + patch[2][1].x + 1) >> 1;
const int t1 = (patch[1][0].w + patch[1][1].y + 1) >> 1;
const int t2 = (patch[0][1].x + patch[0][1].z + patch[2][1].x + patch[2][1].z + 2) >> 2;
const int t3 = (patch[0][1].y + patch[1][1].x + patch[1][1].z + patch[2][1].y + 2) >> 2;
const int t4 = (patch[0][1].z + patch[2][1].z + 1) >> 1;
const int t5 = (patch[1][1].y + patch[1][1].w + 1) >> 1;
const int t6 = (patch[0][1].z + patch[0][2].x + patch[2][1].z + patch[2][2].x + 2) >> 2;
const int t7 = (patch[0][1].w + patch[1][1].z + patch[1][2].x + patch[2][1].w + 2) >> 2;
if ((s_y & 1) ^ blue_last)
{
res0.x = t1;
res0.y = patch[1][1].x;
res0.z = t0;
res1.x = patch[1][1].y;
res1.y = t3;
res1.z = t2;
res2.x = t5;
res2.y = patch[1][1].z;
res2.z = t4;
res3.x = patch[1][1].w;
res3.y = t7;
res3.z = t6;
}
else
{
res0.x = t0;
res0.y = patch[1][1].x;
res0.z = t1;
res1.x = t2;
res1.y = t3;
res1.z = patch[1][1].y;
res2.x = t4;
res2.y = patch[1][1].z;
res2.z = t5;
res3.x = t6;
res3.y = t7;
res3.z = patch[1][1].w;
}
}
else
{
const int t0 = (patch[0][0].w + patch[0][1].y + patch[2][0].w + patch[2][1].y + 2) >> 2;
const int t1 = (patch[0][1].x + patch[1][0].w + patch[1][1].y + patch[2][1].x + 2) >> 2;
const int t2 = (patch[0][1].y + patch[2][1].y + 1) >> 1;
const int t3 = (patch[1][1].x + patch[1][1].z + 1) >> 1;
const int t4 = (patch[0][1].y + patch[0][1].w + patch[2][1].y + patch[2][1].w + 2) >> 2;
const int t5 = (patch[0][1].z + patch[1][1].y + patch[1][1].w + patch[2][1].z + 2) >> 2;
const int t6 = (patch[0][1].w + patch[2][1].w + 1) >> 1;
const int t7 = (patch[1][1].z + patch[1][2].x + 1) >> 1;
if ((s_y & 1) ^ blue_last)
{
res0.x = patch[1][1].x;
res0.y = t1;
res0.z = t0;
res1.x = t3;
res1.y = patch[1][1].y;
res1.z = t2;
res2.x = patch[1][1].z;
res2.y = t5;
res2.z = t4;
res3.x = t7;
res3.y = patch[1][1].w;
res3.z = t6;
}
else
{
res0.x = t0;
res0.y = t1;
res0.z = patch[1][1].x;
res1.x = t2;
res1.y = patch[1][1].y;
res1.z = t3;
res2.x = t4;
res2.y = t5;
res2.z = patch[1][1].z;
res3.x = t6;
res3.y = patch[1][1].w;
res3.z = t7;
}
}
}
};
template <typename D> __device__ __forceinline__ D toDst(const uchar3& pix);
template <> __device__ __forceinline__ uchar toDst<uchar>(const uchar3& pix)
{
typename bgr_to_gray_traits<uchar>::functor_type f = bgr_to_gray_traits<uchar>::create_functor();
return f(pix);
}
template <> __device__ __forceinline__ uchar3 toDst<uchar3>(const uchar3& pix)
{
return pix;
}
template <> __device__ __forceinline__ uchar4 toDst<uchar4>(const uchar3& pix)
{
return make_uchar4(pix.x, pix.y, pix.z, 255);
}
template <typename D>
__global__ void Bayer2BGR_8u(const PtrStepSzb src, PtrStep<D> dst, const bool blue_last, const bool start_with_green)
{
const int s_x = blockIdx.x * blockDim.x + threadIdx.x;
int s_y = blockIdx.y * blockDim.y + threadIdx.y;
if (s_y >= src.rows || (s_x << 2) >= src.cols)
return;
s_y = ::min(::max(s_y, 1), src.rows - 2);
Bayer2BGR<uchar> bayer;
bayer.apply(src, s_x, s_y, blue_last, start_with_green);
const int d_x = (blockIdx.x * blockDim.x + threadIdx.x) << 2;
const int d_y = blockIdx.y * blockDim.y + threadIdx.y;
dst(d_y, d_x) = toDst<D>(bayer.res0);
if (d_x + 1 < src.cols)
dst(d_y, d_x + 1) = toDst<D>(bayer.res1);
if (d_x + 2 < src.cols)
dst(d_y, d_x + 2) = toDst<D>(bayer.res2);
if (d_x + 3 < src.cols)
dst(d_y, d_x + 3) = toDst<D>(bayer.res3);
}
template <> struct Bayer2BGR<ushort>
{
ushort3 res0;
ushort3 res1;
__device__ void apply(const PtrStepSzb& src, int s_x, int s_y, bool blue_last, bool start_with_green)
{
ushort2 patch[3][3];
patch[0][1] = ((const ushort2*) src.ptr(s_y - 1))[s_x];
patch[0][0] = ((const ushort2*) src.ptr(s_y - 1))[::max(s_x - 1, 0)];
patch[0][2] = ((const ushort2*) src.ptr(s_y - 1))[::min(s_x + 1, ((src.cols + 1) >> 1) - 1)];
patch[1][1] = ((const ushort2*) src.ptr(s_y))[s_x];
patch[1][0] = ((const ushort2*) src.ptr(s_y))[::max(s_x - 1, 0)];
patch[1][2] = ((const ushort2*) src.ptr(s_y))[::min(s_x + 1, ((src.cols + 1) >> 1) - 1)];
patch[2][1] = ((const ushort2*) src.ptr(s_y + 1))[s_x];
patch[2][0] = ((const ushort2*) src.ptr(s_y + 1))[::max(s_x - 1, 0)];
patch[2][2] = ((const ushort2*) src.ptr(s_y + 1))[::min(s_x + 1, ((src.cols + 1) >> 1) - 1)];
if ((s_y & 1) ^ start_with_green)
{
const int t0 = (patch[0][1].x + patch[2][1].x + 1) >> 1;
const int t1 = (patch[1][0].y + patch[1][1].y + 1) >> 1;
const int t2 = (patch[0][1].x + patch[0][2].x + patch[2][1].x + patch[2][2].x + 2) >> 2;
const int t3 = (patch[0][1].y + patch[1][1].x + patch[1][2].x + patch[2][1].y + 2) >> 2;
if ((s_y & 1) ^ blue_last)
{
res0.x = t1;
res0.y = patch[1][1].x;
res0.z = t0;
res1.x = patch[1][1].y;
res1.y = t3;
res1.z = t2;
}
else
{
res0.x = t0;
res0.y = patch[1][1].x;
res0.z = t1;
res1.x = t2;
res1.y = t3;
res1.z = patch[1][1].y;
}
}
else
{
const int t0 = (patch[0][0].y + patch[0][1].y + patch[2][0].y + patch[2][1].y + 2) >> 2;
const int t1 = (patch[0][1].x + patch[1][0].y + patch[1][1].y + patch[2][1].x + 2) >> 2;
const int t2 = (patch[0][1].y + patch[2][1].y + 1) >> 1;
const int t3 = (patch[1][1].x + patch[1][2].x + 1) >> 1;
if ((s_y & 1) ^ blue_last)
{
res0.x = patch[1][1].x;
res0.y = t1;
res0.z = t0;
res1.x = t3;
res1.y = patch[1][1].y;
res1.z = t2;
}
else
{
res0.x = t0;
res0.y = t1;
res0.z = patch[1][1].x;
res1.x = t2;
res1.y = patch[1][1].y;
res1.z = t3;
}
}
}
};
template <typename D> __device__ __forceinline__ D toDst(const ushort3& pix);
template <> __device__ __forceinline__ ushort toDst<ushort>(const ushort3& pix)
{
typename bgr_to_gray_traits<ushort>::functor_type f = bgr_to_gray_traits<ushort>::create_functor();
return f(pix);
}
template <> __device__ __forceinline__ ushort3 toDst<ushort3>(const ushort3& pix)
{
return pix;
}
template <> __device__ __forceinline__ ushort4 toDst<ushort4>(const ushort3& pix)
{
return make_ushort4(pix.x, pix.y, pix.z, numeric_limits<ushort>::max());
}
template <typename D>
__global__ void Bayer2BGR_16u(const PtrStepSzb src, PtrStep<D> dst, const bool blue_last, const bool start_with_green)
{
const int s_x = blockIdx.x * blockDim.x + threadIdx.x;
int s_y = blockIdx.y * blockDim.y + threadIdx.y;
if (s_y >= src.rows || (s_x << 1) >= src.cols)
return;
s_y = ::min(::max(s_y, 1), src.rows - 2);
Bayer2BGR<ushort> bayer;
bayer.apply(src, s_x, s_y, blue_last, start_with_green);
const int d_x = (blockIdx.x * blockDim.x + threadIdx.x) << 1;
const int d_y = blockIdx.y * blockDim.y + threadIdx.y;
dst(d_y, d_x) = toDst<D>(bayer.res0);
if (d_x + 1 < src.cols)
dst(d_y, d_x + 1) = toDst<D>(bayer.res1);
}
template <int cn>
void Bayer2BGR_8u_gpu(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream)
{
typedef typename TypeVec<uchar, cn>::vec_type dst_t;
const dim3 block(32, 8);
const dim3 grid(divUp(src.cols, 4 * block.x), divUp(src.rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(Bayer2BGR_8u<dst_t>, cudaFuncCachePreferL1) );
Bayer2BGR_8u<dst_t><<<grid, block, 0, stream>>>(src, (PtrStepSz<dst_t>)dst, blue_last, start_with_green);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template <int cn>
void Bayer2BGR_16u_gpu(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream)
{
typedef typename TypeVec<ushort, cn>::vec_type dst_t;
const dim3 block(32, 8);
const dim3 grid(divUp(src.cols, 2 * block.x), divUp(src.rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(Bayer2BGR_16u<dst_t>, cudaFuncCachePreferL1) );
Bayer2BGR_16u<dst_t><<<grid, block, 0, stream>>>(src, (PtrStepSz<dst_t>)dst, blue_last, start_with_green);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void Bayer2BGR_8u_gpu<1>(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream);
template void Bayer2BGR_8u_gpu<3>(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream);
template void Bayer2BGR_8u_gpu<4>(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream);
template void Bayer2BGR_16u_gpu<1>(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream);
template void Bayer2BGR_16u_gpu<3>(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream);
template void Bayer2BGR_16u_gpu<4>(PtrStepSzb src, PtrStepSzb dst, bool blue_last, bool start_with_green, cudaStream_t stream);
//////////////////////////////////////////////////////////////
// Bayer Demosaicing (Malvar, He, and Cutler)
//
// by Morgan McGuire, Williams College
// http://graphics.cs.williams.edu/papers/BayerJGT09/#shaders
//
// ported to CUDA
texture<uchar, cudaTextureType2D, cudaReadModeElementType> sourceTex(false, cudaFilterModePoint, cudaAddressModeClamp);
template <typename DstType>
__global__ void MHCdemosaic(PtrStepSz<DstType> dst, const int2 sourceOffset, const int2 firstRed)
{
const float kAx = -1.0f / 8.0f, kAy = -1.5f / 8.0f, kAz = 0.5f / 8.0f /*kAw = -1.0f / 8.0f*/;
const float kBx = 2.0f / 8.0f, /*kBy = 0.0f / 8.0f,*/ /*kBz = 0.0f / 8.0f,*/ kBw = 4.0f / 8.0f ;
const float kCx = 4.0f / 8.0f, kCy = 6.0f / 8.0f, kCz = 5.0f / 8.0f /*kCw = 5.0f / 8.0f*/;
const float /*kDx = 0.0f / 8.0f,*/ kDy = 2.0f / 8.0f, kDz = -1.0f / 8.0f /*kDw = -1.0f / 8.0f*/;
const float kEx = -1.0f / 8.0f, kEy = -1.5f / 8.0f, /*kEz = -1.0f / 8.0f,*/ kEw = 0.5f / 8.0f ;
const float kFx = 2.0f / 8.0f, /*kFy = 0.0f / 8.0f,*/ kFz = 4.0f / 8.0f /*kFw = 0.0f / 8.0f*/;
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x == 0 || x >= dst.cols - 1 || y == 0 || y >= dst.rows - 1)
return;
int2 center;
center.x = x + sourceOffset.x;
center.y = y + sourceOffset.y;
int4 xCoord;
xCoord.x = center.x - 2;
xCoord.y = center.x - 1;
xCoord.z = center.x + 1;
xCoord.w = center.x + 2;
int4 yCoord;
yCoord.x = center.y - 2;
yCoord.y = center.y - 1;
yCoord.z = center.y + 1;
yCoord.w = center.y + 2;
float C = tex2D(sourceTex, center.x, center.y); // ( 0, 0)
float4 Dvec;
Dvec.x = tex2D(sourceTex, xCoord.y, yCoord.y); // (-1,-1)
Dvec.y = tex2D(sourceTex, xCoord.y, yCoord.z); // (-1, 1)
Dvec.z = tex2D(sourceTex, xCoord.z, yCoord.y); // ( 1,-1)
Dvec.w = tex2D(sourceTex, xCoord.z, yCoord.z); // ( 1, 1)
float4 value;
value.x = tex2D(sourceTex, center.x, yCoord.x); // ( 0,-2) A0
value.y = tex2D(sourceTex, center.x, yCoord.y); // ( 0,-1) B0
value.z = tex2D(sourceTex, xCoord.x, center.y); // (-2, 0) E0
value.w = tex2D(sourceTex, xCoord.y, center.y); // (-1, 0) F0
// (A0 + A1), (B0 + B1), (E0 + E1), (F0 + F1)
value.x += tex2D(sourceTex, center.x, yCoord.w); // ( 0, 2) A1
value.y += tex2D(sourceTex, center.x, yCoord.z); // ( 0, 1) B1
value.z += tex2D(sourceTex, xCoord.w, center.y); // ( 2, 0) E1
value.w += tex2D(sourceTex, xCoord.z, center.y); // ( 1, 0) F1
float4 PATTERN;
PATTERN.x = kCx * C;
PATTERN.y = kCy * C;
PATTERN.z = kCz * C;
PATTERN.w = PATTERN.z;
float D = Dvec.x + Dvec.y + Dvec.z + Dvec.w;
// There are five filter patterns (identity, cross, checker,
// theta, phi). Precompute the terms from all of them and then
// use swizzles to assign to color channels.
//
// Channel Matches
// x cross (e.g., EE G)
// y checker (e.g., EE B)
// z theta (e.g., EO R)
// w phi (e.g., EO B)
#define A value.x // A0 + A1
#define B value.y // B0 + B1
#define E value.z // E0 + E1
#define F value.w // F0 + F1
float3 temp;
// PATTERN.yzw += (kD.yz * D).xyy;
temp.x = kDy * D;
temp.y = kDz * D;
PATTERN.y += temp.x;
PATTERN.z += temp.y;
PATTERN.w += temp.y;
// PATTERN += (kA.xyz * A).xyzx;
temp.x = kAx * A;
temp.y = kAy * A;
temp.z = kAz * A;
PATTERN.x += temp.x;
PATTERN.y += temp.y;
PATTERN.z += temp.z;
PATTERN.w += temp.x;
// PATTERN += (kE.xyw * E).xyxz;
temp.x = kEx * E;
temp.y = kEy * E;
temp.z = kEw * E;
PATTERN.x += temp.x;
PATTERN.y += temp.y;
PATTERN.z += temp.x;
PATTERN.w += temp.z;
// PATTERN.xw += kB.xw * B;
PATTERN.x += kBx * B;
PATTERN.w += kBw * B;
// PATTERN.xz += kF.xz * F;
PATTERN.x += kFx * F;
PATTERN.z += kFz * F;
// Determine which of four types of pixels we are on.
int2 alternate;
alternate.x = (x + firstRed.x) % 2;
alternate.y = (y + firstRed.y) % 2;
// in BGR sequence;
uchar3 pixelColor =
(alternate.y == 0) ?
((alternate.x == 0) ?
make_uchar3(saturate_cast<uchar>(PATTERN.y), saturate_cast<uchar>(PATTERN.x), saturate_cast<uchar>(C)) :
make_uchar3(saturate_cast<uchar>(PATTERN.w), saturate_cast<uchar>(C), saturate_cast<uchar>(PATTERN.z))) :
((alternate.x == 0) ?
make_uchar3(saturate_cast<uchar>(PATTERN.z), saturate_cast<uchar>(C), saturate_cast<uchar>(PATTERN.w)) :
make_uchar3(saturate_cast<uchar>(C), saturate_cast<uchar>(PATTERN.x), saturate_cast<uchar>(PATTERN.y)));
dst(y, x) = toDst<DstType>(pixelColor);
}
template <int cn>
void MHCdemosaic(PtrStepSzb src, int2 sourceOffset, PtrStepSzb dst, int2 firstRed, cudaStream_t stream)
{
typedef typename TypeVec<uchar, cn>::vec_type dst_t;
const dim3 block(32, 8);
const dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
bindTexture(&sourceTex, src);
MHCdemosaic<dst_t><<<grid, block, 0, stream>>>((PtrStepSz<dst_t>)dst, sourceOffset, firstRed);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void MHCdemosaic<1>(PtrStepSzb src, int2 sourceOffset, PtrStepSzb dst, int2 firstRed, cudaStream_t stream);
template void MHCdemosaic<3>(PtrStepSzb src, int2 sourceOffset, PtrStepSzb dst, int2 firstRed, cudaStream_t stream);
template void MHCdemosaic<4>(PtrStepSzb src, int2 sourceOffset, PtrStepSzb dst, int2 firstRed, cudaStream_t stream);
}}}
#endif /* CUDA_DISABLER */

143
modules/gpuimgproc/src/cuda/gftt.cu
View File

@@ -1,143 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include <thrust/device_ptr.h>
#include <thrust/sort.h>
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/utility.hpp"
namespace cv { namespace gpu { namespace cudev
{
namespace gfft
{
texture<float, cudaTextureType2D, cudaReadModeElementType> eigTex(0, cudaFilterModePoint, cudaAddressModeClamp);
__device__ int g_counter = 0;
template <class Mask> __global__ void findCorners(float threshold, const Mask mask, float2* corners, int max_count, int rows, int cols)
{
const int j = blockIdx.x * blockDim.x + threadIdx.x;
const int i = blockIdx.y * blockDim.y + threadIdx.y;
if (i > 0 && i < rows - 1 && j > 0 && j < cols - 1 && mask(i, j))
{
float val = tex2D(eigTex, j, i);
if (val > threshold)
{
float maxVal = val;
maxVal = ::fmax(tex2D(eigTex, j - 1, i - 1), maxVal);
maxVal = ::fmax(tex2D(eigTex, j , i - 1), maxVal);
maxVal = ::fmax(tex2D(eigTex, j + 1, i - 1), maxVal);
maxVal = ::fmax(tex2D(eigTex, j - 1, i), maxVal);
maxVal = ::fmax(tex2D(eigTex, j + 1, i), maxVal);
maxVal = ::fmax(tex2D(eigTex, j - 1, i + 1), maxVal);
maxVal = ::fmax(tex2D(eigTex, j , i + 1), maxVal);
maxVal = ::fmax(tex2D(eigTex, j + 1, i + 1), maxVal);
if (val == maxVal)
{
const int ind = ::atomicAdd(&g_counter, 1);
if (ind < max_count)
corners[ind] = make_float2(j, i);
}
}
}
}
int findCorners_gpu(PtrStepSzf eig, float threshold, PtrStepSzb mask, float2* corners, int max_count)
{
void* counter_ptr;
cudaSafeCall( cudaGetSymbolAddress(&counter_ptr, g_counter) );
cudaSafeCall( cudaMemset(counter_ptr, 0, sizeof(int)) );
bindTexture(&eigTex, eig);
dim3 block(16, 16);
dim3 grid(divUp(eig.cols, block.x), divUp(eig.rows, block.y));
if (mask.data)
findCorners<<<grid, block>>>(threshold, SingleMask(mask), corners, max_count, eig.rows, eig.cols);
else
findCorners<<<grid, block>>>(threshold, WithOutMask(), corners, max_count, eig.rows, eig.cols);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
int count;
cudaSafeCall( cudaMemcpy(&count, counter_ptr, sizeof(int), cudaMemcpyDeviceToHost) );
return std::min(count, max_count);
}
class EigGreater
{
public:
__device__ __forceinline__ bool operator()(float2 a, float2 b) const
{
return tex2D(eigTex, a.x, a.y) > tex2D(eigTex, b.x, b.y);
}
};
void sortCorners_gpu(PtrStepSzf eig, float2* corners, int count)
{
bindTexture(&eigTex, eig);
thrust::device_ptr<float2> ptr(corners);
thrust::sort(ptr, ptr + count, EigGreater());
}
} // namespace optical_flow
}}}
#endif /* CUDA_DISABLER */

153
modules/gpuimgproc/src/cuda/hist.cu
View File

@@ -1,153 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "opencv2/core/cuda/emulation.hpp"
#include "opencv2/core/cuda/transform.hpp"
using namespace cv::gpu;
using namespace cv::gpu::cudev;
namespace hist
{
__global__ void histogram256Kernel(const uchar* src, int cols, int rows, size_t step, int* hist)
{
__shared__ int shist[256];
const int y = blockIdx.x * blockDim.y + threadIdx.y;
const int tid = threadIdx.y * blockDim.x + threadIdx.x;
shist[tid] = 0;
__syncthreads();
if (y < rows)
{
const unsigned int* rowPtr = (const unsigned int*) (src + y * step);
const int cols_4 = cols / 4;
for (int x = threadIdx.x; x < cols_4; x += blockDim.x)
{
unsigned int data = rowPtr[x];
Emulation::smem::atomicAdd(&shist[(data >> 0) & 0xFFU], 1);
Emulation::smem::atomicAdd(&shist[(data >> 8) & 0xFFU], 1);
Emulation::smem::atomicAdd(&shist[(data >> 16) & 0xFFU], 1);
Emulation::smem::atomicAdd(&shist[(data >> 24) & 0xFFU], 1);
}
if (cols % 4 != 0 && threadIdx.x == 0)
{
for (int x = cols_4 * 4; x < cols; ++x)
{
unsigned int data = ((const uchar*)rowPtr)[x];
Emulation::smem::atomicAdd(&shist[data], 1);
}
}
}
__syncthreads();
const int histVal = shist[tid];
if (histVal > 0)
::atomicAdd(hist + tid, histVal);
}
void histogram256(PtrStepSzb src, int* hist, cudaStream_t stream)
{
const dim3 block(32, 8);
const dim3 grid(divUp(src.rows, block.y));
histogram256Kernel<<<grid, block, 0, stream>>>(src.data, src.cols, src.rows, src.step, hist);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
/////////////////////////////////////////////////////////////////////////
namespace hist
{
__constant__ int c_lut[256];
struct EqualizeHist : unary_function<uchar, uchar>
{
float scale;
__host__ EqualizeHist(float _scale) : scale(_scale) {}
__device__ __forceinline__ uchar operator ()(uchar val) const
{
const int lut = c_lut[val];
return __float2int_rn(scale * lut);
}
};
}
namespace cv { namespace gpu { namespace cudev
{
template <> struct TransformFunctorTraits<hist::EqualizeHist> : DefaultTransformFunctorTraits<hist::EqualizeHist>
{
enum { smart_shift = 4 };
};
}}}
namespace hist
{
void equalizeHist(PtrStepSzb src, PtrStepSzb dst, const int* lut, cudaStream_t stream)
{
if (stream == 0)
cudaSafeCall( cudaMemcpyToSymbol(c_lut, lut, 256 * sizeof(int), 0, cudaMemcpyDeviceToDevice) );
else
cudaSafeCall( cudaMemcpyToSymbolAsync(c_lut, lut, 256 * sizeof(int), 0, cudaMemcpyDeviceToDevice, stream) );
const float scale = 255.0f / (src.cols * src.rows);
cudev::transform(src, dst, EqualizeHist(scale), WithOutMask(), stream);
}
}
#endif /* CUDA_DISABLER */

1709
modules/gpuimgproc/src/cuda/hough.cu
View File

File diff suppressed because it is too large Load Diff

916
modules/gpuimgproc/src/cuda/match_template.cu
View File

@@ -1,916 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_math.hpp"
namespace cv { namespace gpu { namespace cudev
{
namespace match_template
{
__device__ __forceinline__ float sum(float v) { return v; }
__device__ __forceinline__ float sum(float2 v) { return v.x + v.y; }
__device__ __forceinline__ float sum(float3 v) { return v.x + v.y + v.z; }
__device__ __forceinline__ float sum(float4 v) { return v.x + v.y + v.z + v.w; }
__device__ __forceinline__ float first(float v) { return v; }
__device__ __forceinline__ float first(float2 v) { return v.x; }
__device__ __forceinline__ float first(float3 v) { return v.x; }
__device__ __forceinline__ float first(float4 v) { return v.x; }
__device__ __forceinline__ float mul(float a, float b) { return a * b; }
__device__ __forceinline__ float2 mul(float2 a, float2 b) { return make_float2(a.x * b.x, a.y * b.y); }
__device__ __forceinline__ float3 mul(float3 a, float3 b) { return make_float3(a.x * b.x, a.y * b.y, a.z * b.z); }
__device__ __forceinline__ float4 mul(float4 a, float4 b) { return make_float4(a.x * b.x, a.y * b.y, a.z * b.z, a.w * b.w); }
__device__ __forceinline__ float mul(uchar a, uchar b) { return a * b; }
__device__ __forceinline__ float2 mul(uchar2 a, uchar2 b) { return make_float2(a.x * b.x, a.y * b.y); }
__device__ __forceinline__ float3 mul(uchar3 a, uchar3 b) { return make_float3(a.x * b.x, a.y * b.y, a.z * b.z); }
__device__ __forceinline__ float4 mul(uchar4 a, uchar4 b) { return make_float4(a.x * b.x, a.y * b.y, a.z * b.z, a.w * b.w); }
__device__ __forceinline__ float sub(float a, float b) { return a - b; }
__device__ __forceinline__ float2 sub(float2 a, float2 b) { return make_float2(a.x - b.x, a.y - b.y); }
__device__ __forceinline__ float3 sub(float3 a, float3 b) { return make_float3(a.x - b.x, a.y - b.y, a.z - b.z); }
__device__ __forceinline__ float4 sub(float4 a, float4 b) { return make_float4(a.x - b.x, a.y - b.y, a.z - b.z, a.w - b.w); }
__device__ __forceinline__ float sub(uchar a, uchar b) { return a - b; }
__device__ __forceinline__ float2 sub(uchar2 a, uchar2 b) { return make_float2(a.x - b.x, a.y - b.y); }
__device__ __forceinline__ float3 sub(uchar3 a, uchar3 b) { return make_float3(a.x - b.x, a.y - b.y, a.z - b.z); }
__device__ __forceinline__ float4 sub(uchar4 a, uchar4 b) { return make_float4(a.x - b.x, a.y - b.y, a.z - b.z, a.w - b.w); }
//////////////////////////////////////////////////////////////////////
// Naive_CCORR
template <typename T, int cn>
__global__ void matchTemplateNaiveKernel_CCORR(int w, int h, const PtrStepb image, const PtrStepb templ, PtrStepSzf result)
{
typedef typename TypeVec<T, cn>::vec_type Type;
typedef typename TypeVec<float, cn>::vec_type Typef;
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
Typef res = VecTraits<Typef>::all(0);
for (int i = 0; i < h; ++i)
{
const Type* image_ptr = (const Type*)image.ptr(y + i);
const Type* templ_ptr = (const Type*)templ.ptr(i);
for (int j = 0; j < w; ++j)
res = res + mul(image_ptr[x + j], templ_ptr[j]);
}
result.ptr(y)[x] = sum(res);
}
}
template <typename T, int cn>
void matchTemplateNaive_CCORR(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, cudaStream_t stream)
{
const dim3 threads(32, 8);
const dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplateNaiveKernel_CCORR<T, cn><<<grid, threads, 0, stream>>>(templ.cols, templ.rows, image, templ, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void matchTemplateNaive_CCORR_32F(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream)
{
typedef void (*caller_t)(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, cudaStream_t stream);
static const caller_t callers[] =
{
0, matchTemplateNaive_CCORR<float, 1>, matchTemplateNaive_CCORR<float, 2>, matchTemplateNaive_CCORR<float, 3>, matchTemplateNaive_CCORR<float, 4>
};
callers[cn](image, templ, result, stream);
}
void matchTemplateNaive_CCORR_8U(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream)
{
typedef void (*caller_t)(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, cudaStream_t stream);
static const caller_t callers[] =
{
0, matchTemplateNaive_CCORR<uchar, 1>, matchTemplateNaive_CCORR<uchar, 2>, matchTemplateNaive_CCORR<uchar, 3>, matchTemplateNaive_CCORR<uchar, 4>
};
callers[cn](image, templ, result, stream);
}
//////////////////////////////////////////////////////////////////////
// Naive_SQDIFF
template <typename T, int cn>
__global__ void matchTemplateNaiveKernel_SQDIFF(int w, int h, const PtrStepb image, const PtrStepb templ, PtrStepSzf result)
{
typedef typename TypeVec<T, cn>::vec_type Type;
typedef typename TypeVec<float, cn>::vec_type Typef;
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
Typef res = VecTraits<Typef>::all(0);
Typef delta;
for (int i = 0; i < h; ++i)
{
const Type* image_ptr = (const Type*)image.ptr(y + i);
const Type* templ_ptr = (const Type*)templ.ptr(i);
for (int j = 0; j < w; ++j)
{
delta = sub(image_ptr[x + j], templ_ptr[j]);
res = res + delta * delta;
}
}
result.ptr(y)[x] = sum(res);
}
}
template <typename T, int cn>
void matchTemplateNaive_SQDIFF(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, cudaStream_t stream)
{
const dim3 threads(32, 8);
const dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplateNaiveKernel_SQDIFF<T, cn><<<grid, threads, 0, stream>>>(templ.cols, templ.rows, image, templ, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void matchTemplateNaive_SQDIFF_32F(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream)
{
typedef void (*caller_t)(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, cudaStream_t stream);
static const caller_t callers[] =
{
0, matchTemplateNaive_SQDIFF<float, 1>, matchTemplateNaive_SQDIFF<float, 2>, matchTemplateNaive_SQDIFF<float, 3>, matchTemplateNaive_SQDIFF<float, 4>
};
callers[cn](image, templ, result, stream);
}
void matchTemplateNaive_SQDIFF_8U(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream)
{
typedef void (*caller_t)(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, cudaStream_t stream);
static const caller_t callers[] =
{
0, matchTemplateNaive_SQDIFF<uchar, 1>, matchTemplateNaive_SQDIFF<uchar, 2>, matchTemplateNaive_SQDIFF<uchar, 3>, matchTemplateNaive_SQDIFF<uchar, 4>
};
callers[cn](image, templ, result, stream);
}
//////////////////////////////////////////////////////////////////////
// Prepared_SQDIFF
template <int cn>
__global__ void matchTemplatePreparedKernel_SQDIFF_8U(int w, int h, const PtrStep<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sqsum_ = (float)(
(image_sqsum.ptr(y + h)[(x + w) * cn] - image_sqsum.ptr(y)[(x + w) * cn]) -
(image_sqsum.ptr(y + h)[x * cn] - image_sqsum.ptr(y)[x * cn]));
float ccorr = result.ptr(y)[x];
result.ptr(y)[x] = image_sqsum_ - 2.f * ccorr + templ_sqsum;
}
}
template <int cn>
void matchTemplatePrepared_SQDIFF_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result, cudaStream_t stream)
{
const dim3 threads(32, 8);
const dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplatePreparedKernel_SQDIFF_8U<cn><<<grid, threads, 0, stream>>>(w, h, image_sqsum, templ_sqsum, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void matchTemplatePrepared_SQDIFF_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result, int cn,
cudaStream_t stream)
{
typedef void (*caller_t)(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result, cudaStream_t stream);
static const caller_t callers[] =
{
0, matchTemplatePrepared_SQDIFF_8U<1>, matchTemplatePrepared_SQDIFF_8U<2>, matchTemplatePrepared_SQDIFF_8U<3>, matchTemplatePrepared_SQDIFF_8U<4>
};
callers[cn](w, h, image_sqsum, templ_sqsum, result, stream);
}
//////////////////////////////////////////////////////////////////////
// Prepared_SQDIFF_NORMED
// normAcc* are accurate normalization routines which make GPU matchTemplate
// consistent with CPU one
__device__ float normAcc(float num, float denum)
{
if (::fabs(num) < denum)
return num / denum;
if (::fabs(num) < denum * 1.125f)
return num > 0 ? 1 : -1;
return 0;
}
__device__ float normAcc_SQDIFF(float num, float denum)
{
if (::fabs(num) < denum)
return num / denum;
if (::fabs(num) < denum * 1.125f)
return num > 0 ? 1 : -1;
return 1;
}
template <int cn>
__global__ void matchTemplatePreparedKernel_SQDIFF_NORMED_8U(
int w, int h, const PtrStep<unsigned long long> image_sqsum,
unsigned long long templ_sqsum, PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sqsum_ = (float)(
(image_sqsum.ptr(y + h)[(x + w) * cn] - image_sqsum.ptr(y)[(x + w) * cn]) -
(image_sqsum.ptr(y + h)[x * cn] - image_sqsum.ptr(y)[x * cn]));
float ccorr = result.ptr(y)[x];
result.ptr(y)[x] = normAcc_SQDIFF(image_sqsum_ - 2.f * ccorr + templ_sqsum,
sqrtf(image_sqsum_ * templ_sqsum));
}
}
template <int cn>
void matchTemplatePrepared_SQDIFF_NORMED_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum,
PtrStepSzf result, cudaStream_t stream)
{
const dim3 threads(32, 8);
const dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplatePreparedKernel_SQDIFF_NORMED_8U<cn><<<grid, threads, 0, stream>>>(w, h, image_sqsum, templ_sqsum, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void matchTemplatePrepared_SQDIFF_NORMED_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum,
PtrStepSzf result, int cn, cudaStream_t stream)
{
typedef void (*caller_t)(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result, cudaStream_t stream);
static const caller_t callers[] =
{
0, matchTemplatePrepared_SQDIFF_NORMED_8U<1>, matchTemplatePrepared_SQDIFF_NORMED_8U<2>, matchTemplatePrepared_SQDIFF_NORMED_8U<3>, matchTemplatePrepared_SQDIFF_NORMED_8U<4>
};
callers[cn](w, h, image_sqsum, templ_sqsum, result, stream);
}
//////////////////////////////////////////////////////////////////////
// Prepared_CCOFF
__global__ void matchTemplatePreparedKernel_CCOFF_8U(int w, int h, float templ_sum_scale, const PtrStep<unsigned int> image_sum, PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_ = (float)(
(image_sum.ptr(y + h)[x + w] - image_sum.ptr(y)[x + w]) -
(image_sum.ptr(y + h)[x] - image_sum.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
result.ptr(y)[x] = ccorr - image_sum_ * templ_sum_scale;
}
}
void matchTemplatePrepared_CCOFF_8U(int w, int h, const PtrStepSz<unsigned int> image_sum, unsigned int templ_sum, PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplatePreparedKernel_CCOFF_8U<<<grid, threads, 0, stream>>>(w, h, (float)templ_sum / (w * h), image_sum, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void matchTemplatePreparedKernel_CCOFF_8UC2(
int w, int h, float templ_sum_scale_r, float templ_sum_scale_g,
const PtrStep<unsigned int> image_sum_r,
const PtrStep<unsigned int> image_sum_g,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_r_ = (float)(
(image_sum_r.ptr(y + h)[x + w] - image_sum_r.ptr(y)[x + w]) -
(image_sum_r.ptr(y + h)[x] - image_sum_r.ptr(y)[x]));
float image_sum_g_ = (float)(
(image_sum_g.ptr(y + h)[x + w] - image_sum_g.ptr(y)[x + w]) -
(image_sum_g.ptr(y + h)[x] - image_sum_g.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
result.ptr(y)[x] = ccorr - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g;
}
}
void matchTemplatePrepared_CCOFF_8UC2(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r,
const PtrStepSz<unsigned int> image_sum_g,
unsigned int templ_sum_r, unsigned int templ_sum_g,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplatePreparedKernel_CCOFF_8UC2<<<grid, threads, 0, stream>>>(
w, h, (float)templ_sum_r / (w * h), (float)templ_sum_g / (w * h),
image_sum_r, image_sum_g, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void matchTemplatePreparedKernel_CCOFF_8UC3(
int w, int h,
float templ_sum_scale_r,
float templ_sum_scale_g,
float templ_sum_scale_b,
const PtrStep<unsigned int> image_sum_r,
const PtrStep<unsigned int> image_sum_g,
const PtrStep<unsigned int> image_sum_b,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_r_ = (float)(
(image_sum_r.ptr(y + h)[x + w] - image_sum_r.ptr(y)[x + w]) -
(image_sum_r.ptr(y + h)[x] - image_sum_r.ptr(y)[x]));
float image_sum_g_ = (float)(
(image_sum_g.ptr(y + h)[x + w] - image_sum_g.ptr(y)[x + w]) -
(image_sum_g.ptr(y + h)[x] - image_sum_g.ptr(y)[x]));
float image_sum_b_ = (float)(
(image_sum_b.ptr(y + h)[x + w] - image_sum_b.ptr(y)[x + w]) -
(image_sum_b.ptr(y + h)[x] - image_sum_b.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
result.ptr(y)[x] = ccorr - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g
- image_sum_b_ * templ_sum_scale_b;
}
}
void matchTemplatePrepared_CCOFF_8UC3(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r,
const PtrStepSz<unsigned int> image_sum_g,
const PtrStepSz<unsigned int> image_sum_b,
unsigned int templ_sum_r,
unsigned int templ_sum_g,
unsigned int templ_sum_b,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplatePreparedKernel_CCOFF_8UC3<<<grid, threads, 0, stream>>>(
w, h,
(float)templ_sum_r / (w * h),
(float)templ_sum_g / (w * h),
(float)templ_sum_b / (w * h),
image_sum_r, image_sum_g, image_sum_b, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void matchTemplatePreparedKernel_CCOFF_8UC4(
int w, int h,
float templ_sum_scale_r,
float templ_sum_scale_g,
float templ_sum_scale_b,
float templ_sum_scale_a,
const PtrStep<unsigned int> image_sum_r,
const PtrStep<unsigned int> image_sum_g,
const PtrStep<unsigned int> image_sum_b,
const PtrStep<unsigned int> image_sum_a,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_r_ = (float)(
(image_sum_r.ptr(y + h)[x + w] - image_sum_r.ptr(y)[x + w]) -
(image_sum_r.ptr(y + h)[x] - image_sum_r.ptr(y)[x]));
float image_sum_g_ = (float)(
(image_sum_g.ptr(y + h)[x + w] - image_sum_g.ptr(y)[x + w]) -
(image_sum_g.ptr(y + h)[x] - image_sum_g.ptr(y)[x]));
float image_sum_b_ = (float)(
(image_sum_b.ptr(y + h)[x + w] - image_sum_b.ptr(y)[x + w]) -
(image_sum_b.ptr(y + h)[x] - image_sum_b.ptr(y)[x]));
float image_sum_a_ = (float)(
(image_sum_a.ptr(y + h)[x + w] - image_sum_a.ptr(y)[x + w]) -
(image_sum_a.ptr(y + h)[x] - image_sum_a.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
result.ptr(y)[x] = ccorr - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g
- image_sum_b_ * templ_sum_scale_b
- image_sum_a_ * templ_sum_scale_a;
}
}
void matchTemplatePrepared_CCOFF_8UC4(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r,
const PtrStepSz<unsigned int> image_sum_g,
const PtrStepSz<unsigned int> image_sum_b,
const PtrStepSz<unsigned int> image_sum_a,
unsigned int templ_sum_r,
unsigned int templ_sum_g,
unsigned int templ_sum_b,
unsigned int templ_sum_a,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
matchTemplatePreparedKernel_CCOFF_8UC4<<<grid, threads, 0, stream>>>(
w, h,
(float)templ_sum_r / (w * h),
(float)templ_sum_g / (w * h),
(float)templ_sum_b / (w * h),
(float)templ_sum_a / (w * h),
image_sum_r, image_sum_g, image_sum_b, image_sum_a,
result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
//////////////////////////////////////////////////////////////////////
// Prepared_CCOFF_NORMED
__global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8U(
int w, int h, float weight,
float templ_sum_scale, float templ_sqsum_scale,
const PtrStep<unsigned int> image_sum,
const PtrStep<unsigned long long> image_sqsum,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float ccorr = result.ptr(y)[x];
float image_sum_ = (float)(
(image_sum.ptr(y + h)[x + w] - image_sum.ptr(y)[x + w]) -
(image_sum.ptr(y + h)[x] - image_sum.ptr(y)[x]));
float image_sqsum_ = (float)(
(image_sqsum.ptr(y + h)[x + w] - image_sqsum.ptr(y)[x + w]) -
(image_sqsum.ptr(y + h)[x] - image_sqsum.ptr(y)[x]));
result.ptr(y)[x] = normAcc(ccorr - image_sum_ * templ_sum_scale,
sqrtf(templ_sqsum_scale * (image_sqsum_ - weight * image_sum_ * image_sum_)));
}
}
void matchTemplatePrepared_CCOFF_NORMED_8U(
int w, int h, const PtrStepSz<unsigned int> image_sum,
const PtrStepSz<unsigned long long> image_sqsum,
unsigned int templ_sum, unsigned long long templ_sqsum,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
float weight = 1.f / (w * h);
float templ_sum_scale = templ_sum * weight;
float templ_sqsum_scale = templ_sqsum - weight * templ_sum * templ_sum;
matchTemplatePreparedKernel_CCOFF_NORMED_8U<<<grid, threads, 0, stream>>>(
w, h, weight, templ_sum_scale, templ_sqsum_scale,
image_sum, image_sqsum, result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8UC2(
int w, int h, float weight,
float templ_sum_scale_r, float templ_sum_scale_g,
float templ_sqsum_scale,
const PtrStep<unsigned int> image_sum_r, const PtrStep<unsigned long long> image_sqsum_r,
const PtrStep<unsigned int> image_sum_g, const PtrStep<unsigned long long> image_sqsum_g,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_r_ = (float)(
(image_sum_r.ptr(y + h)[x + w] - image_sum_r.ptr(y)[x + w]) -
(image_sum_r.ptr(y + h)[x] - image_sum_r.ptr(y)[x]));
float image_sqsum_r_ = (float)(
(image_sqsum_r.ptr(y + h)[x + w] - image_sqsum_r.ptr(y)[x + w]) -
(image_sqsum_r.ptr(y + h)[x] - image_sqsum_r.ptr(y)[x]));
float image_sum_g_ = (float)(
(image_sum_g.ptr(y + h)[x + w] - image_sum_g.ptr(y)[x + w]) -
(image_sum_g.ptr(y + h)[x] - image_sum_g.ptr(y)[x]));
float image_sqsum_g_ = (float)(
(image_sqsum_g.ptr(y + h)[x + w] - image_sqsum_g.ptr(y)[x + w]) -
(image_sqsum_g.ptr(y + h)[x] - image_sqsum_g.ptr(y)[x]));
float num = result.ptr(y)[x] - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g;
float denum = sqrtf(templ_sqsum_scale * (image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_));
result.ptr(y)[x] = normAcc(num, denum);
}
}
void matchTemplatePrepared_CCOFF_NORMED_8UC2(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r, const PtrStepSz<unsigned long long> image_sqsum_r,
const PtrStepSz<unsigned int> image_sum_g, const PtrStepSz<unsigned long long> image_sqsum_g,
unsigned int templ_sum_r, unsigned long long templ_sqsum_r,
unsigned int templ_sum_g, unsigned long long templ_sqsum_g,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
float weight = 1.f / (w * h);
float templ_sum_scale_r = templ_sum_r * weight;
float templ_sum_scale_g = templ_sum_g * weight;
float templ_sqsum_scale = templ_sqsum_r - weight * templ_sum_r * templ_sum_r
+ templ_sqsum_g - weight * templ_sum_g * templ_sum_g;
matchTemplatePreparedKernel_CCOFF_NORMED_8UC2<<<grid, threads, 0, stream>>>(
w, h, weight,
templ_sum_scale_r, templ_sum_scale_g,
templ_sqsum_scale,
image_sum_r, image_sqsum_r,
image_sum_g, image_sqsum_g,
result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8UC3(
int w, int h, float weight,
float templ_sum_scale_r, float templ_sum_scale_g, float templ_sum_scale_b,
float templ_sqsum_scale,
const PtrStep<unsigned int> image_sum_r, const PtrStep<unsigned long long> image_sqsum_r,
const PtrStep<unsigned int> image_sum_g, const PtrStep<unsigned long long> image_sqsum_g,
const PtrStep<unsigned int> image_sum_b, const PtrStep<unsigned long long> image_sqsum_b,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_r_ = (float)(
(image_sum_r.ptr(y + h)[x + w] - image_sum_r.ptr(y)[x + w]) -
(image_sum_r.ptr(y + h)[x] - image_sum_r.ptr(y)[x]));
float image_sqsum_r_ = (float)(
(image_sqsum_r.ptr(y + h)[x + w] - image_sqsum_r.ptr(y)[x + w]) -
(image_sqsum_r.ptr(y + h)[x] - image_sqsum_r.ptr(y)[x]));
float image_sum_g_ = (float)(
(image_sum_g.ptr(y + h)[x + w] - image_sum_g.ptr(y)[x + w]) -
(image_sum_g.ptr(y + h)[x] - image_sum_g.ptr(y)[x]));
float image_sqsum_g_ = (float)(
(image_sqsum_g.ptr(y + h)[x + w] - image_sqsum_g.ptr(y)[x + w]) -
(image_sqsum_g.ptr(y + h)[x] - image_sqsum_g.ptr(y)[x]));
float image_sum_b_ = (float)(
(image_sum_b.ptr(y + h)[x + w] - image_sum_b.ptr(y)[x + w]) -
(image_sum_b.ptr(y + h)[x] - image_sum_b.ptr(y)[x]));
float image_sqsum_b_ = (float)(
(image_sqsum_b.ptr(y + h)[x + w] - image_sqsum_b.ptr(y)[x + w]) -
(image_sqsum_b.ptr(y + h)[x] - image_sqsum_b.ptr(y)[x]));
float num = result.ptr(y)[x] - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g
- image_sum_b_ * templ_sum_scale_b;
float denum = sqrtf(templ_sqsum_scale * (image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_
+ image_sqsum_b_ - weight * image_sum_b_ * image_sum_b_));
result.ptr(y)[x] = normAcc(num, denum);
}
}
void matchTemplatePrepared_CCOFF_NORMED_8UC3(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r, const PtrStepSz<unsigned long long> image_sqsum_r,
const PtrStepSz<unsigned int> image_sum_g, const PtrStepSz<unsigned long long> image_sqsum_g,
const PtrStepSz<unsigned int> image_sum_b, const PtrStepSz<unsigned long long> image_sqsum_b,
unsigned int templ_sum_r, unsigned long long templ_sqsum_r,
unsigned int templ_sum_g, unsigned long long templ_sqsum_g,
unsigned int templ_sum_b, unsigned long long templ_sqsum_b,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
float weight = 1.f / (w * h);
float templ_sum_scale_r = templ_sum_r * weight;
float templ_sum_scale_g = templ_sum_g * weight;
float templ_sum_scale_b = templ_sum_b * weight;
float templ_sqsum_scale = templ_sqsum_r - weight * templ_sum_r * templ_sum_r
+ templ_sqsum_g - weight * templ_sum_g * templ_sum_g
+ templ_sqsum_b - weight * templ_sum_b * templ_sum_b;
matchTemplatePreparedKernel_CCOFF_NORMED_8UC3<<<grid, threads, 0, stream>>>(
w, h, weight,
templ_sum_scale_r, templ_sum_scale_g, templ_sum_scale_b,
templ_sqsum_scale,
image_sum_r, image_sqsum_r,
image_sum_g, image_sqsum_g,
image_sum_b, image_sqsum_b,
result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8UC4(
int w, int h, float weight,
float templ_sum_scale_r, float templ_sum_scale_g, float templ_sum_scale_b,
float templ_sum_scale_a, float templ_sqsum_scale,
const PtrStep<unsigned int> image_sum_r, const PtrStep<unsigned long long> image_sqsum_r,
const PtrStep<unsigned int> image_sum_g, const PtrStep<unsigned long long> image_sqsum_g,
const PtrStep<unsigned int> image_sum_b, const PtrStep<unsigned long long> image_sqsum_b,
const PtrStep<unsigned int> image_sum_a, const PtrStep<unsigned long long> image_sqsum_a,
PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sum_r_ = (float)(
(image_sum_r.ptr(y + h)[x + w] - image_sum_r.ptr(y)[x + w]) -
(image_sum_r.ptr(y + h)[x] - image_sum_r.ptr(y)[x]));
float image_sqsum_r_ = (float)(
(image_sqsum_r.ptr(y + h)[x + w] - image_sqsum_r.ptr(y)[x + w]) -
(image_sqsum_r.ptr(y + h)[x] - image_sqsum_r.ptr(y)[x]));
float image_sum_g_ = (float)(
(image_sum_g.ptr(y + h)[x + w] - image_sum_g.ptr(y)[x + w]) -
(image_sum_g.ptr(y + h)[x] - image_sum_g.ptr(y)[x]));
float image_sqsum_g_ = (float)(
(image_sqsum_g.ptr(y + h)[x + w] - image_sqsum_g.ptr(y)[x + w]) -
(image_sqsum_g.ptr(y + h)[x] - image_sqsum_g.ptr(y)[x]));
float image_sum_b_ = (float)(
(image_sum_b.ptr(y + h)[x + w] - image_sum_b.ptr(y)[x + w]) -
(image_sum_b.ptr(y + h)[x] - image_sum_b.ptr(y)[x]));
float image_sqsum_b_ = (float)(
(image_sqsum_b.ptr(y + h)[x + w] - image_sqsum_b.ptr(y)[x + w]) -
(image_sqsum_b.ptr(y + h)[x] - image_sqsum_b.ptr(y)[x]));
float image_sum_a_ = (float)(
(image_sum_a.ptr(y + h)[x + w] - image_sum_a.ptr(y)[x + w]) -
(image_sum_a.ptr(y + h)[x] - image_sum_a.ptr(y)[x]));
float image_sqsum_a_ = (float)(
(image_sqsum_a.ptr(y + h)[x + w] - image_sqsum_a.ptr(y)[x + w]) -
(image_sqsum_a.ptr(y + h)[x] - image_sqsum_a.ptr(y)[x]));
float num = result.ptr(y)[x] - image_sum_r_ * templ_sum_scale_r - image_sum_g_ * templ_sum_scale_g
- image_sum_b_ * templ_sum_scale_b - image_sum_a_ * templ_sum_scale_a;
float denum = sqrtf(templ_sqsum_scale * (image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_
+ image_sqsum_b_ - weight * image_sum_b_ * image_sum_b_
+ image_sqsum_a_ - weight * image_sum_a_ * image_sum_a_));
result.ptr(y)[x] = normAcc(num, denum);
}
}
void matchTemplatePrepared_CCOFF_NORMED_8UC4(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r, const PtrStepSz<unsigned long long> image_sqsum_r,
const PtrStepSz<unsigned int> image_sum_g, const PtrStepSz<unsigned long long> image_sqsum_g,
const PtrStepSz<unsigned int> image_sum_b, const PtrStepSz<unsigned long long> image_sqsum_b,
const PtrStepSz<unsigned int> image_sum_a, const PtrStepSz<unsigned long long> image_sqsum_a,
unsigned int templ_sum_r, unsigned long long templ_sqsum_r,
unsigned int templ_sum_g, unsigned long long templ_sqsum_g,
unsigned int templ_sum_b, unsigned long long templ_sqsum_b,
unsigned int templ_sum_a, unsigned long long templ_sqsum_a,
PtrStepSzf result, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
float weight = 1.f / (w * h);
float templ_sum_scale_r = templ_sum_r * weight;
float templ_sum_scale_g = templ_sum_g * weight;
float templ_sum_scale_b = templ_sum_b * weight;
float templ_sum_scale_a = templ_sum_a * weight;
float templ_sqsum_scale = templ_sqsum_r - weight * templ_sum_r * templ_sum_r
+ templ_sqsum_g - weight * templ_sum_g * templ_sum_g
+ templ_sqsum_b - weight * templ_sum_b * templ_sum_b
+ templ_sqsum_a - weight * templ_sum_a * templ_sum_a;
matchTemplatePreparedKernel_CCOFF_NORMED_8UC4<<<grid, threads, 0, stream>>>(
w, h, weight,
templ_sum_scale_r, templ_sum_scale_g, templ_sum_scale_b, templ_sum_scale_a,
templ_sqsum_scale,
image_sum_r, image_sqsum_r,
image_sum_g, image_sqsum_g,
image_sum_b, image_sqsum_b,
image_sum_a, image_sqsum_a,
result);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
//////////////////////////////////////////////////////////////////////
// normalize
template <int cn>
__global__ void normalizeKernel_8U(
int w, int h, const PtrStep<unsigned long long> image_sqsum,
unsigned long long templ_sqsum, PtrStepSzf result)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
float image_sqsum_ = (float)(
(image_sqsum.ptr(y + h)[(x + w) * cn] - image_sqsum.ptr(y)[(x + w) * cn]) -
(image_sqsum.ptr(y + h)[x * cn] - image_sqsum.ptr(y)[x * cn]));
result.ptr(y)[x] = normAcc(result.ptr(y)[x], sqrtf(image_sqsum_ * templ_sqsum));
}
}
void normalize_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum,
unsigned long long templ_sqsum, PtrStepSzf result, int cn, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
switch (cn)
{
case 1:
normalizeKernel_8U<1><<<grid, threads, 0, stream>>>(w, h, image_sqsum, templ_sqsum, result);
break;
case 2:
normalizeKernel_8U<2><<<grid, threads, 0, stream>>>(w, h, image_sqsum, templ_sqsum, result);
break;
case 3:
normalizeKernel_8U<3><<<grid, threads, 0, stream>>>(w, h, image_sqsum, templ_sqsum, result);
break;
case 4:
normalizeKernel_8U<4><<<grid, threads, 0, stream>>>(w, h, image_sqsum, templ_sqsum, result);
break;
}
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
//////////////////////////////////////////////////////////////////////
// extractFirstChannel
template <int cn>
__global__ void extractFirstChannel_32F(const PtrStepb image, PtrStepSzf result)
{
typedef typename TypeVec<float, cn>::vec_type Typef;
int x = blockDim.x * blockIdx.x + threadIdx.x;
int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < result.cols && y < result.rows)
{
Typef val = ((const Typef*)image.ptr(y))[x];
result.ptr(y)[x] = first(val);
}
}
void extractFirstChannel_32F(const PtrStepSzb image, PtrStepSzf result, int cn, cudaStream_t stream)
{
dim3 threads(32, 8);
dim3 grid(divUp(result.cols, threads.x), divUp(result.rows, threads.y));
switch (cn)
{
case 1:
extractFirstChannel_32F<1><<<grid, threads, 0, stream>>>(image, result);
break;
case 2:
extractFirstChannel_32F<2><<<grid, threads, 0, stream>>>(image, result);
break;
case 3:
extractFirstChannel_32F<3><<<grid, threads, 0, stream>>>(image, result);
break;
case 4:
extractFirstChannel_32F<4><<<grid, threads, 0, stream>>>(image, result);
break;
}
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
} //namespace match_template
}}} // namespace cv { namespace gpu { namespace cudev
#endif /* CUDA_DISABLER */

182
modules/gpuimgproc/src/cuda/mean_shift.cu
View File

@@ -1,182 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_traits.hpp"
#include "opencv2/core/cuda/vec_math.hpp"
#include "opencv2/core/cuda/saturate_cast.hpp"
#include "opencv2/core/cuda/border_interpolate.hpp"
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
texture<uchar4, 2> tex_meanshift;
__device__ short2 do_mean_shift(int x0, int y0, unsigned char* out,
size_t out_step, int cols, int rows,
int sp, int sr, int maxIter, float eps)
{
int isr2 = sr*sr;
uchar4 c = tex2D(tex_meanshift, x0, y0 );
// iterate meanshift procedure
for( int iter = 0; iter < maxIter; iter++ )
{
int count = 0;
int s0 = 0, s1 = 0, s2 = 0, sx = 0, sy = 0;
float icount;
//mean shift: process pixels in window (p-sigmaSp)x(p+sigmaSp)
int minx = x0-sp;
int miny = y0-sp;
int maxx = x0+sp;
int maxy = y0+sp;
for( int y = miny; y <= maxy; y++)
{
int rowCount = 0;
for( int x = minx; x <= maxx; x++ )
{
uchar4 t = tex2D( tex_meanshift, x, y );
int norm2 = (t.x - c.x) * (t.x - c.x) + (t.y - c.y) * (t.y - c.y) + (t.z - c.z) * (t.z - c.z);
if( norm2 <= isr2 )
{
s0 += t.x; s1 += t.y; s2 += t.z;
sx += x; rowCount++;
}
}
count += rowCount;
sy += y*rowCount;
}
if( count == 0 )
break;
icount = 1.f/count;
int x1 = __float2int_rz(sx*icount);
int y1 = __float2int_rz(sy*icount);
s0 = __float2int_rz(s0*icount);
s1 = __float2int_rz(s1*icount);
s2 = __float2int_rz(s2*icount);
int norm2 = (s0 - c.x) * (s0 - c.x) + (s1 - c.y) * (s1 - c.y) + (s2 - c.z) * (s2 - c.z);
bool stopFlag = (x0 == x1 && y0 == y1) || (::abs(x1-x0) + ::abs(y1-y0) + norm2 <= eps);
x0 = x1; y0 = y1;
c.x = s0; c.y = s1; c.z = s2;
if( stopFlag )
break;
}
int base = (blockIdx.y * blockDim.y + threadIdx.y) * out_step + (blockIdx.x * blockDim.x + threadIdx.x) * 4 * sizeof(uchar);
*(uchar4*)(out + base) = c;
return make_short2((short)x0, (short)y0);
}
__global__ void meanshift_kernel(unsigned char* out, size_t out_step, int cols, int rows, int sp, int sr, int maxIter, float eps )
{
int x0 = blockIdx.x * blockDim.x + threadIdx.x;
int y0 = blockIdx.y * blockDim.y + threadIdx.y;
if( x0 < cols && y0 < rows )
do_mean_shift(x0, y0, out, out_step, cols, rows, sp, sr, maxIter, eps);
}
void meanShiftFiltering_gpu(const PtrStepSzb& src, PtrStepSzb dst, int sp, int sr, int maxIter, float eps, cudaStream_t stream)
{
dim3 grid(1, 1, 1);
dim3 threads(32, 8, 1);
grid.x = divUp(src.cols, threads.x);
grid.y = divUp(src.rows, threads.y);
cudaChannelFormatDesc desc = cudaCreateChannelDesc<uchar4>();
cudaSafeCall( cudaBindTexture2D( 0, tex_meanshift, src.data, desc, src.cols, src.rows, src.step ) );
meanshift_kernel<<< grid, threads, 0, stream >>>( dst.data, dst.step, dst.cols, dst.rows, sp, sr, maxIter, eps );
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void meanshiftproc_kernel(unsigned char* outr, size_t outrstep,
unsigned char* outsp, size_t outspstep,
int cols, int rows,
int sp, int sr, int maxIter, float eps)
{
int x0 = blockIdx.x * blockDim.x + threadIdx.x;
int y0 = blockIdx.y * blockDim.y + threadIdx.y;
if( x0 < cols && y0 < rows )
{
int basesp = (blockIdx.y * blockDim.y + threadIdx.y) * outspstep + (blockIdx.x * blockDim.x + threadIdx.x) * 2 * sizeof(short);
*(short2*)(outsp + basesp) = do_mean_shift(x0, y0, outr, outrstep, cols, rows, sp, sr, maxIter, eps);
}
}
void meanShiftProc_gpu(const PtrStepSzb& src, PtrStepSzb dstr, PtrStepSzb dstsp, int sp, int sr, int maxIter, float eps, cudaStream_t stream)
{
dim3 grid(1, 1, 1);
dim3 threads(32, 8, 1);
grid.x = divUp(src.cols, threads.x);
grid.y = divUp(src.rows, threads.y);
cudaChannelFormatDesc desc = cudaCreateChannelDesc<uchar4>();
cudaSafeCall( cudaBindTexture2D( 0, tex_meanshift, src.data, desc, src.cols, src.rows, src.step ) );
meanshiftproc_kernel<<< grid, threads, 0, stream >>>( dstr.data, dstr.step, dstsp.data, dstsp.step, dstr.cols, dstr.rows, sp, sr, maxIter, eps );
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
}}}
#endif

274
modules/gpuimgproc/src/cvt_color_internal.h
View File

@@ -1,274 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __cvt_color_internal_h__
#define __cvt_color_internal_h__
namespace cv { namespace gpu { namespace cudev
{
#define OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name) \
void name(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
#define OPENCV_GPU_DECLARE_CVTCOLOR_ALL(name) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _8u) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _16u) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _32f)
#define OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(name) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _8u) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _32f)
#define OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(name) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _8u) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _32f) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _full_8u) \
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(name ## _full_32f)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr_to_bgr555)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr_to_bgr565)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(rgb_to_bgr555)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(rgb_to_bgr565)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgra_to_bgr555)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgra_to_bgr565)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(rgba_to_bgr555)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(rgba_to_bgr565)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr555_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr565_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr555_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr565_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr555_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr565_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr555_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr565_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(gray_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(gray_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(gray_to_bgr555)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(gray_to_bgr565)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr555_to_gray)
OPENCV_GPU_DECLARE_CVTCOLOR_ONE(bgr565_to_gray)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_gray)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_gray)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_gray)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_gray)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_yuv)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_yuv)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_yuv4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_yuv4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_yuv)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_yuv)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_yuv4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_yuv4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(yuv4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_YCrCb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_YCrCb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_YCrCb4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_YCrCb4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_YCrCb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_YCrCb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_YCrCb4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_YCrCb4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(YCrCb4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_xyz)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_xyz)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgb_to_xyz4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(rgba_to_xyz4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_xyz)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_xyz)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgr_to_xyz4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(bgra_to_xyz4)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_ALL(xyz4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgb_to_hsv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgba_to_hsv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgb_to_hsv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgba_to_hsv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgr_to_hsv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgra_to_hsv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgr_to_hsv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgra_to_hsv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hsv4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgb_to_hls)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgba_to_hls)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgb_to_hls4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(rgba_to_hls4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgr_to_hls)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgra_to_hls)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgr_to_hls4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(bgra_to_hls4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL(hls4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgb_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgba_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgb_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgba_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgr_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgra_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgr_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgra_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgb_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgba_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgb_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgba_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgr_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgra_to_lab)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgr_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgra_to_lab4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_lrgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_lrgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_lrgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_lrgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_lbgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_lbgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab_to_lbgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lab4_to_lbgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgb_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgba_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgb_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(rgba_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgr_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgra_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgr_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(bgra_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgb_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgba_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgb_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lrgba_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgr_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgra_to_luv)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgr_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(lbgra_to_luv4)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_rgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_rgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_bgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_bgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_lrgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_lrgb)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_lrgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_lrgba)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_lbgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_lbgr)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv_to_lbgra)
OPENCV_GPU_DECLARE_CVTCOLOR_8U32F(luv4_to_lbgra)
#undef OPENCV_GPU_DECLARE_CVTCOLOR_ONE
#undef OPENCV_GPU_DECLARE_CVTCOLOR_ALL
#undef OPENCV_GPU_DECLARE_CVTCOLOR_8U32F
#undef OPENCV_GPU_DECLARE_CVTCOLOR_8U32F_FULL
}}}
#endif

176
modules/gpuimgproc/src/gftt.cpp
View File

@@ -1,176 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::GoodFeaturesToTrackDetector_GPU::operator ()(const GpuMat&, GpuMat&, const GpuMat&) { throw_no_cuda(); }
#else /* !defined (HAVE_CUDA) */
namespace cv { namespace gpu { namespace cudev
{
namespace gfft
{
int findCorners_gpu(PtrStepSzf eig, float threshold, PtrStepSzb mask, float2* corners, int max_count);
void sortCorners_gpu(PtrStepSzf eig, float2* corners, int count);
}
}}}
void cv::gpu::GoodFeaturesToTrackDetector_GPU::operator ()(const GpuMat& image, GpuMat& corners, const GpuMat& mask)
{
#ifndef HAVE_OPENCV_GPUARITHM
(void) image;
(void) corners;
(void) mask;
throw_no_cuda();
#else
using namespace cv::gpu::cudev::gfft;
CV_Assert(qualityLevel > 0 && minDistance >= 0 && maxCorners >= 0);
CV_Assert(mask.empty() || (mask.type() == CV_8UC1 && mask.size() == image.size()));
ensureSizeIsEnough(image.size(), CV_32F, eig_);
if (useHarrisDetector)
cornerHarris(image, eig_, Dx_, Dy_, buf_, blockSize, 3, harrisK);
else
cornerMinEigenVal(image, eig_, Dx_, Dy_, buf_, blockSize, 3);
double maxVal = 0;
gpu::minMax(eig_, 0, &maxVal, GpuMat(), minMaxbuf_);
ensureSizeIsEnough(1, std::max(1000, static_cast<int>(image.size().area() * 0.05)), CV_32FC2, tmpCorners_);
int total = findCorners_gpu(eig_, static_cast<float>(maxVal * qualityLevel), mask, tmpCorners_.ptr<float2>(), tmpCorners_.cols);
if (total == 0)
{
corners.release();
return;
}
sortCorners_gpu(eig_, tmpCorners_.ptr<float2>(), total);
if (minDistance < 1)
tmpCorners_.colRange(0, maxCorners > 0 ? std::min(maxCorners, total) : total).copyTo(corners);
else
{
std::vector<Point2f> tmp(total);
Mat tmpMat(1, total, CV_32FC2, (void*)&tmp[0]);
tmpCorners_.colRange(0, total).download(tmpMat);
std::vector<Point2f> tmp2;
tmp2.reserve(total);
const int cell_size = cvRound(minDistance);
const int grid_width = (image.cols + cell_size - 1) / cell_size;
const int grid_height = (image.rows + cell_size - 1) / cell_size;
std::vector< std::vector<Point2f> > grid(grid_width * grid_height);
for (int i = 0; i < total; ++i)
{
Point2f p = tmp[i];
bool good = true;
int x_cell = static_cast<int>(p.x / cell_size);
int y_cell = static_cast<int>(p.y / cell_size);
int x1 = x_cell - 1;
int y1 = y_cell - 1;
int x2 = x_cell + 1;
int y2 = y_cell + 1;
// boundary check
x1 = std::max(0, x1);
y1 = std::max(0, y1);
x2 = std::min(grid_width - 1, x2);
y2 = std::min(grid_height - 1, y2);
for (int yy = y1; yy <= y2; yy++)
{
for (int xx = x1; xx <= x2; xx++)
{
std::vector<Point2f>& m = grid[yy * grid_width + xx];
if (!m.empty())
{
for(size_t j = 0; j < m.size(); j++)
{
float dx = p.x - m[j].x;
float dy = p.y - m[j].y;
if (dx * dx + dy * dy < minDistance * minDistance)
{
good = false;
goto break_out;
}
}
}
}
}
break_out:
if(good)
{
grid[y_cell * grid_width + x_cell].push_back(p);
tmp2.push_back(p);
if (maxCorners > 0 && tmp2.size() == static_cast<size_t>(maxCorners))
break;
}
}
corners.upload(Mat(1, static_cast<int>(tmp2.size()), CV_32FC2, &tmp2[0]));
}
#endif
}
#endif /* !defined (HAVE_CUDA) */

557
modules/gpuimgproc/src/histogram.cpp
View File

@@ -1,557 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::evenLevels(GpuMat&, int, int, int) { throw_no_cuda(); }
void cv::gpu::histEven(const GpuMat&, GpuMat&, int, int, int, Stream&) { throw_no_cuda(); }
void cv::gpu::histEven(const GpuMat&, GpuMat&, GpuMat&, int, int, int, Stream&) { throw_no_cuda(); }
void cv::gpu::histEven(const GpuMat&, GpuMat*, int*, int*, int*, Stream&) { throw_no_cuda(); }
void cv::gpu::histEven(const GpuMat&, GpuMat*, GpuMat&, int*, int*, int*, Stream&) { throw_no_cuda(); }
void cv::gpu::histRange(const GpuMat&, GpuMat&, const GpuMat&, Stream&) { throw_no_cuda(); }
void cv::gpu::histRange(const GpuMat&, GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::gpu::histRange(const GpuMat&, GpuMat*, const GpuMat*, Stream&) { throw_no_cuda(); }
void cv::gpu::histRange(const GpuMat&, GpuMat*, const GpuMat*, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::gpu::calcHist(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::gpu::equalizeHist(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::gpu::equalizeHist(const GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
cv::Ptr<cv::gpu::CLAHE> cv::gpu::createCLAHE(double, cv::Size) { throw_no_cuda(); return cv::Ptr<cv::gpu::CLAHE>(); }
#else /* !defined (HAVE_CUDA) */
////////////////////////////////////////////////////////////////////////
// NPP Histogram
namespace
{
typedef NppStatus (*get_buf_size_c1_t)(NppiSize oSizeROI, int nLevels, int* hpBufferSize);
typedef NppStatus (*get_buf_size_c4_t)(NppiSize oSizeROI, int nLevels[], int* hpBufferSize);
template<int SDEPTH> struct NppHistogramEvenFuncC1
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef NppStatus (*func_ptr)(const src_t* pSrc, int nSrcStep, NppiSize oSizeROI, Npp32s * pHist,
int nLevels, Npp32s nLowerLevel, Npp32s nUpperLevel, Npp8u * pBuffer);
};
template<int SDEPTH> struct NppHistogramEvenFuncC4
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef NppStatus (*func_ptr)(const src_t* pSrc, int nSrcStep, NppiSize oSizeROI,
Npp32s * pHist[4], int nLevels[4], Npp32s nLowerLevel[4], Npp32s nUpperLevel[4], Npp8u * pBuffer);
};
template<int SDEPTH, typename NppHistogramEvenFuncC1<SDEPTH>::func_ptr func, get_buf_size_c1_t get_buf_size>
struct NppHistogramEvenC1
{
typedef typename NppHistogramEvenFuncC1<SDEPTH>::src_t src_t;
static void hist(const GpuMat& src, GpuMat& hist, GpuMat& buffer, int histSize, int lowerLevel, int upperLevel, cudaStream_t stream)
{
int levels = histSize + 1;
hist.create(1, histSize, CV_32S);
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
int buf_size;
get_buf_size(sz, levels, &buf_size);
ensureSizeIsEnough(1, buf_size, CV_8U, buffer);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<src_t>(), static_cast<int>(src.step), sz, hist.ptr<Npp32s>(), levels,
lowerLevel, upperLevel, buffer.ptr<Npp8u>()) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int SDEPTH, typename NppHistogramEvenFuncC4<SDEPTH>::func_ptr func, get_buf_size_c4_t get_buf_size>
struct NppHistogramEvenC4
{
typedef typename NppHistogramEvenFuncC4<SDEPTH>::src_t src_t;
static void hist(const GpuMat& src, GpuMat hist[4], GpuMat& buffer, int histSize[4], int lowerLevel[4], int upperLevel[4], cudaStream_t stream)
{
int levels[] = {histSize[0] + 1, histSize[1] + 1, histSize[2] + 1, histSize[3] + 1};
hist[0].create(1, histSize[0], CV_32S);
hist[1].create(1, histSize[1], CV_32S);
hist[2].create(1, histSize[2], CV_32S);
hist[3].create(1, histSize[3], CV_32S);
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
Npp32s* pHist[] = {hist[0].ptr<Npp32s>(), hist[1].ptr<Npp32s>(), hist[2].ptr<Npp32s>(), hist[3].ptr<Npp32s>()};
int buf_size;
get_buf_size(sz, levels, &buf_size);
ensureSizeIsEnough(1, buf_size, CV_8U, buffer);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<src_t>(), static_cast<int>(src.step), sz, pHist, levels, lowerLevel, upperLevel, buffer.ptr<Npp8u>()) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int SDEPTH> struct NppHistogramRangeFuncC1
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef Npp32s level_t;
enum {LEVEL_TYPE_CODE=CV_32SC1};
typedef NppStatus (*func_ptr)(const src_t* pSrc, int nSrcStep, NppiSize oSizeROI, Npp32s* pHist,
const Npp32s* pLevels, int nLevels, Npp8u* pBuffer);
};
template<> struct NppHistogramRangeFuncC1<CV_32F>
{
typedef Npp32f src_t;
typedef Npp32f level_t;
enum {LEVEL_TYPE_CODE=CV_32FC1};
typedef NppStatus (*func_ptr)(const Npp32f* pSrc, int nSrcStep, NppiSize oSizeROI, Npp32s* pHist,
const Npp32f* pLevels, int nLevels, Npp8u* pBuffer);
};
template<int SDEPTH> struct NppHistogramRangeFuncC4
{
typedef typename NPPTypeTraits<SDEPTH>::npp_type src_t;
typedef Npp32s level_t;
enum {LEVEL_TYPE_CODE=CV_32SC1};
typedef NppStatus (*func_ptr)(const src_t* pSrc, int nSrcStep, NppiSize oSizeROI, Npp32s* pHist[4],
const Npp32s* pLevels[4], int nLevels[4], Npp8u* pBuffer);
};
template<> struct NppHistogramRangeFuncC4<CV_32F>
{
typedef Npp32f src_t;
typedef Npp32f level_t;
enum {LEVEL_TYPE_CODE=CV_32FC1};
typedef NppStatus (*func_ptr)(const Npp32f* pSrc, int nSrcStep, NppiSize oSizeROI, Npp32s* pHist[4],
const Npp32f* pLevels[4], int nLevels[4], Npp8u* pBuffer);
};
template<int SDEPTH, typename NppHistogramRangeFuncC1<SDEPTH>::func_ptr func, get_buf_size_c1_t get_buf_size>
struct NppHistogramRangeC1
{
typedef typename NppHistogramRangeFuncC1<SDEPTH>::src_t src_t;
typedef typename NppHistogramRangeFuncC1<SDEPTH>::level_t level_t;
enum {LEVEL_TYPE_CODE=NppHistogramRangeFuncC1<SDEPTH>::LEVEL_TYPE_CODE};
static void hist(const GpuMat& src, GpuMat& hist, const GpuMat& levels, GpuMat& buffer, cudaStream_t stream)
{
CV_Assert(levels.type() == LEVEL_TYPE_CODE && levels.rows == 1);
hist.create(1, levels.cols - 1, CV_32S);
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
int buf_size;
get_buf_size(sz, levels.cols, &buf_size);
ensureSizeIsEnough(1, buf_size, CV_8U, buffer);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<src_t>(), static_cast<int>(src.step), sz, hist.ptr<Npp32s>(), levels.ptr<level_t>(), levels.cols, buffer.ptr<Npp8u>()) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<int SDEPTH, typename NppHistogramRangeFuncC4<SDEPTH>::func_ptr func, get_buf_size_c4_t get_buf_size>
struct NppHistogramRangeC4
{
typedef typename NppHistogramRangeFuncC4<SDEPTH>::src_t src_t;
typedef typename NppHistogramRangeFuncC1<SDEPTH>::level_t level_t;
enum {LEVEL_TYPE_CODE=NppHistogramRangeFuncC1<SDEPTH>::LEVEL_TYPE_CODE};
static void hist(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4], GpuMat& buffer, cudaStream_t stream)
{
CV_Assert(levels[0].type() == LEVEL_TYPE_CODE && levels[0].rows == 1);
CV_Assert(levels[1].type() == LEVEL_TYPE_CODE && levels[1].rows == 1);
CV_Assert(levels[2].type() == LEVEL_TYPE_CODE && levels[2].rows == 1);
CV_Assert(levels[3].type() == LEVEL_TYPE_CODE && levels[3].rows == 1);
hist[0].create(1, levels[0].cols - 1, CV_32S);
hist[1].create(1, levels[1].cols - 1, CV_32S);
hist[2].create(1, levels[2].cols - 1, CV_32S);
hist[3].create(1, levels[3].cols - 1, CV_32S);
Npp32s* pHist[] = {hist[0].ptr<Npp32s>(), hist[1].ptr<Npp32s>(), hist[2].ptr<Npp32s>(), hist[3].ptr<Npp32s>()};
int nLevels[] = {levels[0].cols, levels[1].cols, levels[2].cols, levels[3].cols};
const level_t* pLevels[] = {levels[0].ptr<level_t>(), levels[1].ptr<level_t>(), levels[2].ptr<level_t>(), levels[3].ptr<level_t>()};
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
int buf_size;
get_buf_size(sz, nLevels, &buf_size);
ensureSizeIsEnough(1, buf_size, CV_8U, buffer);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<src_t>(), static_cast<int>(src.step), sz, pHist, pLevels, nLevels, buffer.ptr<Npp8u>()) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
};
}
void cv::gpu::evenLevels(GpuMat& levels, int nLevels, int lowerLevel, int upperLevel)
{
Mat host_levels(1, nLevels, CV_32SC1);
nppSafeCall( nppiEvenLevelsHost_32s(host_levels.ptr<Npp32s>(), nLevels, lowerLevel, upperLevel) );
levels.upload(host_levels);
}
void cv::gpu::histEven(const GpuMat& src, GpuMat& hist, int histSize, int lowerLevel, int upperLevel, Stream& stream)
{
GpuMat buf;
histEven(src, hist, buf, histSize, lowerLevel, upperLevel, stream);
}
void cv::gpu::histEven(const GpuMat& src, GpuMat& hist, GpuMat& buf, int histSize, int lowerLevel, int upperLevel, Stream& stream)
{
CV_Assert(src.type() == CV_8UC1 || src.type() == CV_16UC1 || src.type() == CV_16SC1 );
typedef void (*hist_t)(const GpuMat& src, GpuMat& hist, GpuMat& buf, int levels, int lowerLevel, int upperLevel, cudaStream_t stream);
static const hist_t hist_callers[] =
{
NppHistogramEvenC1<CV_8U , nppiHistogramEven_8u_C1R , nppiHistogramEvenGetBufferSize_8u_C1R >::hist,
0,
NppHistogramEvenC1<CV_16U, nppiHistogramEven_16u_C1R, nppiHistogramEvenGetBufferSize_16u_C1R>::hist,
NppHistogramEvenC1<CV_16S, nppiHistogramEven_16s_C1R, nppiHistogramEvenGetBufferSize_16s_C1R>::hist
};
hist_callers[src.depth()](src, hist, buf, histSize, lowerLevel, upperLevel, StreamAccessor::getStream(stream));
}
void cv::gpu::histEven(const GpuMat& src, GpuMat hist[4], int histSize[4], int lowerLevel[4], int upperLevel[4], Stream& stream)
{
GpuMat buf;
histEven(src, hist, buf, histSize, lowerLevel, upperLevel, stream);
}
void cv::gpu::histEven(const GpuMat& src, GpuMat hist[4], GpuMat& buf, int histSize[4], int lowerLevel[4], int upperLevel[4], Stream& stream)
{
CV_Assert(src.type() == CV_8UC4 || src.type() == CV_16UC4 || src.type() == CV_16SC4 );
typedef void (*hist_t)(const GpuMat& src, GpuMat hist[4], GpuMat& buf, int levels[4], int lowerLevel[4], int upperLevel[4], cudaStream_t stream);
static const hist_t hist_callers[] =
{
NppHistogramEvenC4<CV_8U , nppiHistogramEven_8u_C4R , nppiHistogramEvenGetBufferSize_8u_C4R >::hist,
0,
NppHistogramEvenC4<CV_16U, nppiHistogramEven_16u_C4R, nppiHistogramEvenGetBufferSize_16u_C4R>::hist,
NppHistogramEvenC4<CV_16S, nppiHistogramEven_16s_C4R, nppiHistogramEvenGetBufferSize_16s_C4R>::hist
};
hist_callers[src.depth()](src, hist, buf, histSize, lowerLevel, upperLevel, StreamAccessor::getStream(stream));
}
void cv::gpu::histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels, Stream& stream)
{
GpuMat buf;
histRange(src, hist, levels, buf, stream);
}
void cv::gpu::histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels, GpuMat& buf, Stream& stream)
{
CV_Assert(src.type() == CV_8UC1 || src.type() == CV_16UC1 || src.type() == CV_16SC1 || src.type() == CV_32FC1);
typedef void (*hist_t)(const GpuMat& src, GpuMat& hist, const GpuMat& levels, GpuMat& buf, cudaStream_t stream);
static const hist_t hist_callers[] =
{
NppHistogramRangeC1<CV_8U , nppiHistogramRange_8u_C1R , nppiHistogramRangeGetBufferSize_8u_C1R >::hist,
0,
NppHistogramRangeC1<CV_16U, nppiHistogramRange_16u_C1R, nppiHistogramRangeGetBufferSize_16u_C1R>::hist,
NppHistogramRangeC1<CV_16S, nppiHistogramRange_16s_C1R, nppiHistogramRangeGetBufferSize_16s_C1R>::hist,
0,
NppHistogramRangeC1<CV_32F, nppiHistogramRange_32f_C1R, nppiHistogramRangeGetBufferSize_32f_C1R>::hist
};
hist_callers[src.depth()](src, hist, levels, buf, StreamAccessor::getStream(stream));
}
void cv::gpu::histRange(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4], Stream& stream)
{
GpuMat buf;
histRange(src, hist, levels, buf, stream);
}
void cv::gpu::histRange(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4], GpuMat& buf, Stream& stream)
{
CV_Assert(src.type() == CV_8UC4 || src.type() == CV_16UC4 || src.type() == CV_16SC4 || src.type() == CV_32FC4);
typedef void (*hist_t)(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4], GpuMat& buf, cudaStream_t stream);
static const hist_t hist_callers[] =
{
NppHistogramRangeC4<CV_8U , nppiHistogramRange_8u_C4R , nppiHistogramRangeGetBufferSize_8u_C4R >::hist,
0,
NppHistogramRangeC4<CV_16U, nppiHistogramRange_16u_C4R, nppiHistogramRangeGetBufferSize_16u_C4R>::hist,
NppHistogramRangeC4<CV_16S, nppiHistogramRange_16s_C4R, nppiHistogramRangeGetBufferSize_16s_C4R>::hist,
0,
NppHistogramRangeC4<CV_32F, nppiHistogramRange_32f_C4R, nppiHistogramRangeGetBufferSize_32f_C4R>::hist
};
hist_callers[src.depth()](src, hist, levels, buf, StreamAccessor::getStream(stream));
}
////////////////////////////////////////////////////////////////////////
// calcHist
namespace hist
{
void histogram256(PtrStepSzb src, int* hist, cudaStream_t stream);
}
void cv::gpu::calcHist(const GpuMat& src, GpuMat& hist, Stream& stream)
{
CV_Assert(src.type() == CV_8UC1);
hist.create(1, 256, CV_32SC1);
hist.setTo(Scalar::all(0));
hist::histogram256(src, hist.ptr<int>(), StreamAccessor::getStream(stream));
}
////////////////////////////////////////////////////////////////////////
// equalizeHist
namespace hist
{
void equalizeHist(PtrStepSzb src, PtrStepSzb dst, const int* lut, cudaStream_t stream);
}
void cv::gpu::equalizeHist(const GpuMat& src, GpuMat& dst, Stream& stream)
{
GpuMat hist;
GpuMat buf;
equalizeHist(src, dst, hist, buf, stream);
}
void cv::gpu::equalizeHist(const GpuMat& src, GpuMat& dst, GpuMat& hist, GpuMat& buf, Stream& s)
{
CV_Assert(src.type() == CV_8UC1);
dst.create(src.size(), src.type());
int intBufSize;
nppSafeCall( nppsIntegralGetBufferSize_32s(256, &intBufSize) );
ensureSizeIsEnough(1, intBufSize + 256 * sizeof(int), CV_8UC1, buf);
GpuMat intBuf(1, intBufSize, CV_8UC1, buf.ptr());
GpuMat lut(1, 256, CV_32S, buf.ptr() + intBufSize);
calcHist(src, hist, s);
cudaStream_t stream = StreamAccessor::getStream(s);
NppStreamHandler h(stream);
nppSafeCall( nppsIntegral_32s(hist.ptr<Npp32s>(), lut.ptr<Npp32s>(), 256, intBuf.ptr<Npp8u>()) );
hist::equalizeHist(src, dst, lut.ptr<int>(), stream);
}
////////////////////////////////////////////////////////////////////////
// CLAHE
namespace clahe
{
void calcLut(PtrStepSzb src, PtrStepb lut, int tilesX, int tilesY, int2 tileSize, int clipLimit, float lutScale, cudaStream_t stream);
void transform(PtrStepSzb src, PtrStepSzb dst, PtrStepb lut, int tilesX, int tilesY, int2 tileSize, cudaStream_t stream);
}
namespace
{
class CLAHE_Impl : public cv::gpu::CLAHE
{
public:
CLAHE_Impl(double clipLimit = 40.0, int tilesX = 8, int tilesY = 8);
cv::AlgorithmInfo* info() const;
void apply(cv::InputArray src, cv::OutputArray dst);
void apply(InputArray src, OutputArray dst, Stream& stream);
void setClipLimit(double clipLimit);
double getClipLimit() const;
void setTilesGridSize(cv::Size tileGridSize);
cv::Size getTilesGridSize() const;
void collectGarbage();
private:
double clipLimit_;
int tilesX_;
int tilesY_;
GpuMat srcExt_;
GpuMat lut_;
};
CLAHE_Impl::CLAHE_Impl(double clipLimit, int tilesX, int tilesY) :
clipLimit_(clipLimit), tilesX_(tilesX), tilesY_(tilesY)
{
}
CV_INIT_ALGORITHM(CLAHE_Impl, "CLAHE_GPU",
obj.info()->addParam(obj, "clipLimit", obj.clipLimit_);
obj.info()->addParam(obj, "tilesX", obj.tilesX_);
obj.info()->addParam(obj, "tilesY", obj.tilesY_))
void CLAHE_Impl::apply(cv::InputArray _src, cv::OutputArray _dst)
{
apply(_src, _dst, Stream::Null());
}
void CLAHE_Impl::apply(InputArray _src, OutputArray _dst, Stream& s)
{
GpuMat src = _src.getGpuMat();
CV_Assert( src.type() == CV_8UC1 );
_dst.create( src.size(), src.type() );
GpuMat dst = _dst.getGpuMat();
const int histSize = 256;
ensureSizeIsEnough(tilesX_ * tilesY_, histSize, CV_8UC1, lut_);
cudaStream_t stream = StreamAccessor::getStream(s);
cv::Size tileSize;
GpuMat srcForLut;
if (src.cols % tilesX_ == 0 && src.rows % tilesY_ == 0)
{
tileSize = cv::Size(src.cols / tilesX_, src.rows / tilesY_);
srcForLut = src;
}
else
{
#ifndef HAVE_OPENCV_GPUARITHM
throw_no_cuda();
#else
cv::gpu::copyMakeBorder(src, srcExt_, 0, tilesY_ - (src.rows % tilesY_), 0, tilesX_ - (src.cols % tilesX_), cv::BORDER_REFLECT_101, cv::Scalar(), s);
#endif
tileSize = cv::Size(srcExt_.cols / tilesX_, srcExt_.rows / tilesY_);
srcForLut = srcExt_;
}
const int tileSizeTotal = tileSize.area();
const float lutScale = static_cast<float>(histSize - 1) / tileSizeTotal;
int clipLimit = 0;
if (clipLimit_ > 0.0)
{
clipLimit = static_cast<int>(clipLimit_ * tileSizeTotal / histSize);
clipLimit = std::max(clipLimit, 1);
}
clahe::calcLut(srcForLut, lut_, tilesX_, tilesY_, make_int2(tileSize.width, tileSize.height), clipLimit, lutScale, stream);
clahe::transform(src, dst, lut_, tilesX_, tilesY_, make_int2(tileSize.width, tileSize.height), stream);
}
void CLAHE_Impl::setClipLimit(double clipLimit)
{
clipLimit_ = clipLimit;
}
double CLAHE_Impl::getClipLimit() const
{
return clipLimit_;
}
void CLAHE_Impl::setTilesGridSize(cv::Size tileGridSize)
{
tilesX_ = tileGridSize.width;
tilesY_ = tileGridSize.height;
}
cv::Size CLAHE_Impl::getTilesGridSize() const
{
return cv::Size(tilesX_, tilesY_);
}
void CLAHE_Impl::collectGarbage()
{
srcExt_.release();
lut_.release();
}
}
cv::Ptr<cv::gpu::CLAHE> cv::gpu::createCLAHE(double clipLimit, cv::Size tileGridSize)
{
return new CLAHE_Impl(clipLimit, tileGridSize.width, tileGridSize.height);
}
#endif /* !defined (HAVE_CUDA) */

1432
modules/gpuimgproc/src/hough.cpp
View File

File diff suppressed because it is too large Load Diff

439
modules/gpuimgproc/src/match_template.cpp
View File

@@ -1,439 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || !defined (HAVE_OPENCV_GPUARITHM) || defined (CUDA_DISABLER)
void cv::gpu::matchTemplate(const GpuMat&, const GpuMat&, GpuMat&, int, Stream&) { throw_no_cuda(); }
#else
namespace cv { namespace gpu { namespace cudev
{
namespace match_template
{
void matchTemplateNaive_CCORR_8U(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream);
void matchTemplateNaive_CCORR_32F(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream);
void matchTemplateNaive_SQDIFF_8U(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream);
void matchTemplateNaive_SQDIFF_32F(const PtrStepSzb image, const PtrStepSzb templ, PtrStepSzf result, int cn, cudaStream_t stream);
void matchTemplatePrepared_SQDIFF_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result,
int cn, cudaStream_t stream);
void matchTemplatePrepared_SQDIFF_NORMED_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum, unsigned long long templ_sqsum, PtrStepSzf result,
int cn, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_8U(int w, int h, const PtrStepSz<unsigned int> image_sum, unsigned int templ_sum, PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_8UC2(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r,
const PtrStepSz<unsigned int> image_sum_g,
unsigned int templ_sum_r,
unsigned int templ_sum_g,
PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_8UC3(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r,
const PtrStepSz<unsigned int> image_sum_g,
const PtrStepSz<unsigned int> image_sum_b,
unsigned int templ_sum_r,
unsigned int templ_sum_g,
unsigned int templ_sum_b,
PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_8UC4(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r,
const PtrStepSz<unsigned int> image_sum_g,
const PtrStepSz<unsigned int> image_sum_b,
const PtrStepSz<unsigned int> image_sum_a,
unsigned int templ_sum_r,
unsigned int templ_sum_g,
unsigned int templ_sum_b,
unsigned int templ_sum_a,
PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_NORMED_8U(
int w, int h, const PtrStepSz<unsigned int> image_sum,
const PtrStepSz<unsigned long long> image_sqsum,
unsigned int templ_sum, unsigned long long templ_sqsum,
PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_NORMED_8UC2(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r, const PtrStepSz<unsigned long long> image_sqsum_r,
const PtrStepSz<unsigned int> image_sum_g, const PtrStepSz<unsigned long long> image_sqsum_g,
unsigned int templ_sum_r, unsigned long long templ_sqsum_r,
unsigned int templ_sum_g, unsigned long long templ_sqsum_g,
PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_NORMED_8UC3(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r, const PtrStepSz<unsigned long long> image_sqsum_r,
const PtrStepSz<unsigned int> image_sum_g, const PtrStepSz<unsigned long long> image_sqsum_g,
const PtrStepSz<unsigned int> image_sum_b, const PtrStepSz<unsigned long long> image_sqsum_b,
unsigned int templ_sum_r, unsigned long long templ_sqsum_r,
unsigned int templ_sum_g, unsigned long long templ_sqsum_g,
unsigned int templ_sum_b, unsigned long long templ_sqsum_b,
PtrStepSzf result, cudaStream_t stream);
void matchTemplatePrepared_CCOFF_NORMED_8UC4(
int w, int h,
const PtrStepSz<unsigned int> image_sum_r, const PtrStepSz<unsigned long long> image_sqsum_r,
const PtrStepSz<unsigned int> image_sum_g, const PtrStepSz<unsigned long long> image_sqsum_g,
const PtrStepSz<unsigned int> image_sum_b, const PtrStepSz<unsigned long long> image_sqsum_b,
const PtrStepSz<unsigned int> image_sum_a, const PtrStepSz<unsigned long long> image_sqsum_a,
unsigned int templ_sum_r, unsigned long long templ_sqsum_r,
unsigned int templ_sum_g, unsigned long long templ_sqsum_g,
unsigned int templ_sum_b, unsigned long long templ_sqsum_b,
unsigned int templ_sum_a, unsigned long long templ_sqsum_a,
PtrStepSzf result, cudaStream_t stream);
void normalize_8U(int w, int h, const PtrStepSz<unsigned long long> image_sqsum,
unsigned long long templ_sqsum, PtrStepSzf result, int cn, cudaStream_t stream);
void extractFirstChannel_32F(const PtrStepSzb image, PtrStepSzf result, int cn, cudaStream_t stream);
}
}}}
using namespace ::cv::gpu::cudev::match_template;
namespace
{
// Evaluates optimal template's area threshold. If
// template's area is less than the threshold, we use naive match
// template version, otherwise FFT-based (if available)
int getTemplateThreshold(int method, int depth)
{
switch (method)
{
case cv::TM_CCORR:
if (depth == CV_32F) return 250;
if (depth == CV_8U) return 300;
break;
case cv::TM_SQDIFF:
if (depth == CV_8U) return 300;
break;
}
CV_Error(cv::Error::StsBadArg, "getTemplateThreshold: unsupported match template mode");
return 0;
}
void matchTemplate_CCORR_32F(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
if (templ.size().area() < getTemplateThreshold(cv::TM_CCORR, CV_32F))
{
matchTemplateNaive_CCORR_32F(image, templ, result, image.channels(), StreamAccessor::getStream(stream));
return;
}
gpu::ConvolveBuf convolve_buf;
convolve_buf.user_block_size = buf.user_block_size;
if (image.channels() == 1)
gpu::convolve(image.reshape(1), templ.reshape(1), result, true, convolve_buf, stream);
else
{
GpuMat result_;
gpu::convolve(image.reshape(1), templ.reshape(1), result_, true, convolve_buf, stream);
extractFirstChannel_32F(result_, result, image.channels(), StreamAccessor::getStream(stream));
}
}
void matchTemplate_CCORR_8U(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
if (templ.size().area() < getTemplateThreshold(cv::TM_CCORR, CV_8U))
{
result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
matchTemplateNaive_CCORR_8U(image, templ, result, image.channels(), StreamAccessor::getStream(stream));
return;
}
if (stream)
{
stream.enqueueConvert(image, buf.imagef, CV_32F);
stream.enqueueConvert(templ, buf.templf, CV_32F);
}
else
{
image.convertTo(buf.imagef, CV_32F);
templ.convertTo(buf.templf, CV_32F);
}
matchTemplate_CCORR_32F(buf.imagef, buf.templf, result, buf, stream);
}
void matchTemplate_CCORR_NORMED_8U(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
matchTemplate_CCORR_8U(image, templ, result, buf, stream);
buf.image_sqsums.resize(1);
gpu::sqrIntegral(image.reshape(1), buf.image_sqsums[0], stream);
unsigned long long templ_sqsum = (unsigned long long)gpu::sqrSum(templ.reshape(1))[0];
normalize_8U(templ.cols, templ.rows, buf.image_sqsums[0], templ_sqsum, result, image.channels(), StreamAccessor::getStream(stream));
}
void matchTemplate_SQDIFF_32F(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
(void)buf;
result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
matchTemplateNaive_SQDIFF_32F(image, templ, result, image.channels(), StreamAccessor::getStream(stream));
}
void matchTemplate_SQDIFF_8U(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
if (templ.size().area() < getTemplateThreshold(cv::TM_SQDIFF, CV_8U))
{
result.create(image.rows - templ.rows + 1, image.cols - templ.cols + 1, CV_32F);
matchTemplateNaive_SQDIFF_8U(image, templ, result, image.channels(), StreamAccessor::getStream(stream));
return;
}
buf.image_sqsums.resize(1);
gpu::sqrIntegral(image.reshape(1), buf.image_sqsums[0], stream);
unsigned long long templ_sqsum = (unsigned long long)gpu::sqrSum(templ.reshape(1))[0];
matchTemplate_CCORR_8U(image, templ, result, buf, stream);
matchTemplatePrepared_SQDIFF_8U(templ.cols, templ.rows, buf.image_sqsums[0], templ_sqsum, result, image.channels(), StreamAccessor::getStream(stream));
}
void matchTemplate_SQDIFF_NORMED_8U(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
buf.image_sqsums.resize(1);
gpu::sqrIntegral(image.reshape(1), buf.image_sqsums[0], stream);
unsigned long long templ_sqsum = (unsigned long long)gpu::sqrSum(templ.reshape(1))[0];
matchTemplate_CCORR_8U(image, templ, result, buf, stream);
matchTemplatePrepared_SQDIFF_NORMED_8U(templ.cols, templ.rows, buf.image_sqsums[0], templ_sqsum, result, image.channels(), StreamAccessor::getStream(stream));
}
void matchTemplate_CCOFF_8U(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
matchTemplate_CCORR_8U(image, templ, result, buf, stream);
if (image.channels() == 1)
{
buf.image_sums.resize(1);
gpu::integral(image, buf.image_sums[0], stream);
unsigned int templ_sum = (unsigned int)sum(templ)[0];
matchTemplatePrepared_CCOFF_8U(templ.cols, templ.rows, buf.image_sums[0], templ_sum, result, StreamAccessor::getStream(stream));
}
else
{
gpu::split(image, buf.images);
buf.image_sums.resize(buf.images.size());
for (int i = 0; i < image.channels(); ++i)
gpu::integral(buf.images[i], buf.image_sums[i], stream);
Scalar templ_sum = gpu::sum(templ);
switch (image.channels())
{
case 2:
matchTemplatePrepared_CCOFF_8UC2(
templ.cols, templ.rows, buf.image_sums[0], buf.image_sums[1],
(unsigned int)templ_sum[0], (unsigned int)templ_sum[1],
result, StreamAccessor::getStream(stream));
break;
case 3:
matchTemplatePrepared_CCOFF_8UC3(
templ.cols, templ.rows, buf.image_sums[0], buf.image_sums[1], buf.image_sums[2],
(unsigned int)templ_sum[0], (unsigned int)templ_sum[1], (unsigned int)templ_sum[2],
result, StreamAccessor::getStream(stream));
break;
case 4:
matchTemplatePrepared_CCOFF_8UC4(
templ.cols, templ.rows, buf.image_sums[0], buf.image_sums[1], buf.image_sums[2], buf.image_sums[3],
(unsigned int)templ_sum[0], (unsigned int)templ_sum[1], (unsigned int)templ_sum[2],
(unsigned int)templ_sum[3], result, StreamAccessor::getStream(stream));
break;
default:
CV_Error(cv::Error::StsBadArg, "matchTemplate: unsupported number of channels");
}
}
}
void matchTemplate_CCOFF_NORMED_8U(
const GpuMat& image, const GpuMat& templ, GpuMat& result, MatchTemplateBuf &buf, Stream& stream)
{
if (stream)
{
stream.enqueueConvert(image, buf.imagef, CV_32F);
stream.enqueueConvert(templ, buf.templf, CV_32F);
}
else
{
image.convertTo(buf.imagef, CV_32F);
templ.convertTo(buf.templf, CV_32F);
}
matchTemplate_CCORR_32F(buf.imagef, buf.templf, result, buf, stream);
if (image.channels() == 1)
{
buf.image_sums.resize(1);
gpu::integral(image, buf.image_sums[0], stream);
buf.image_sqsums.resize(1);
gpu::sqrIntegral(image, buf.image_sqsums[0], stream);
unsigned int templ_sum = (unsigned int)gpu::sum(templ)[0];
unsigned long long templ_sqsum = (unsigned long long)gpu::sqrSum(templ)[0];
matchTemplatePrepared_CCOFF_NORMED_8U(
templ.cols, templ.rows, buf.image_sums[0], buf.image_sqsums[0],
templ_sum, templ_sqsum, result, StreamAccessor::getStream(stream));
}
else
{
gpu::split(image, buf.images);
buf.image_sums.resize(buf.images.size());
buf.image_sqsums.resize(buf.images.size());
for (int i = 0; i < image.channels(); ++i)
{
gpu::integral(buf.images[i], buf.image_sums[i], stream);
gpu::sqrIntegral(buf.images[i], buf.image_sqsums[i], stream);
}
Scalar templ_sum = gpu::sum(templ);
Scalar templ_sqsum = gpu::sqrSum(templ);
switch (image.channels())
{
case 2:
matchTemplatePrepared_CCOFF_NORMED_8UC2(
templ.cols, templ.rows,
buf.image_sums[0], buf.image_sqsums[0],
buf.image_sums[1], buf.image_sqsums[1],
(unsigned int)templ_sum[0], (unsigned long long)templ_sqsum[0],
(unsigned int)templ_sum[1], (unsigned long long)templ_sqsum[1],
result, StreamAccessor::getStream(stream));
break;
case 3:
matchTemplatePrepared_CCOFF_NORMED_8UC3(
templ.cols, templ.rows,
buf.image_sums[0], buf.image_sqsums[0],
buf.image_sums[1], buf.image_sqsums[1],
buf.image_sums[2], buf.image_sqsums[2],
(unsigned int)templ_sum[0], (unsigned long long)templ_sqsum[0],
(unsigned int)templ_sum[1], (unsigned long long)templ_sqsum[1],
(unsigned int)templ_sum[2], (unsigned long long)templ_sqsum[2],
result, StreamAccessor::getStream(stream));
break;
case 4:
matchTemplatePrepared_CCOFF_NORMED_8UC4(
templ.cols, templ.rows,
buf.image_sums[0], buf.image_sqsums[0],
buf.image_sums[1], buf.image_sqsums[1],
buf.image_sums[2], buf.image_sqsums[2],
buf.image_sums[3], buf.image_sqsums[3],
(unsigned int)templ_sum[0], (unsigned long long)templ_sqsum[0],
(unsigned int)templ_sum[1], (unsigned long long)templ_sqsum[1],
(unsigned int)templ_sum[2], (unsigned long long)templ_sqsum[2],
(unsigned int)templ_sum[3], (unsigned long long)templ_sqsum[3],
result, StreamAccessor::getStream(stream));
break;
default:
CV_Error(cv::Error::StsBadArg, "matchTemplate: unsupported number of channels");
}
}
}
}
void cv::gpu::matchTemplate(const GpuMat& image, const GpuMat& templ, GpuMat& result, int method, Stream& stream)
{
MatchTemplateBuf buf;
matchTemplate(image, templ, result, method, buf, stream);
}
void cv::gpu::matchTemplate(
const GpuMat& image, const GpuMat& templ, GpuMat& result, int method,
MatchTemplateBuf &buf, Stream& stream)
{
CV_Assert(image.type() == templ.type());
CV_Assert(image.cols >= templ.cols && image.rows >= templ.rows);
typedef void (*Caller)(const GpuMat&, const GpuMat&, GpuMat&, MatchTemplateBuf&, Stream& stream);
static const Caller callers8U[] = { ::matchTemplate_SQDIFF_8U, ::matchTemplate_SQDIFF_NORMED_8U,
::matchTemplate_CCORR_8U, ::matchTemplate_CCORR_NORMED_8U,
::matchTemplate_CCOFF_8U, ::matchTemplate_CCOFF_NORMED_8U };
static const Caller callers32F[] = { ::matchTemplate_SQDIFF_32F, 0,
::matchTemplate_CCORR_32F, 0, 0, 0 };
const Caller* callers = 0;
switch (image.depth())
{
case CV_8U: callers = callers8U; break;
case CV_32F: callers = callers32F; break;
default: CV_Error(cv::Error::StsBadArg, "matchTemplate: unsupported data type");
}
Caller caller = callers[method];
CV_Assert(caller);
caller(image, templ, result, buf, stream);
}
#endif

128
modules/gpuimgproc/src/mean_shift.cpp
View File

@@ -1,128 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::gpu;
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
void cv::gpu::meanShiftFiltering(const GpuMat&, GpuMat&, int, int, TermCriteria, Stream&) { throw_no_cuda(); }
void cv::gpu::meanShiftProc(const GpuMat&, GpuMat&, GpuMat&, int, int, TermCriteria, Stream&) { throw_no_cuda(); }
#else /* !defined (HAVE_CUDA) */
////////////////////////////////////////////////////////////////////////
// meanShiftFiltering_GPU
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
void meanShiftFiltering_gpu(const PtrStepSzb& src, PtrStepSzb dst, int sp, int sr, int maxIter, float eps, cudaStream_t stream);
}
}}}
void cv::gpu::meanShiftFiltering(const GpuMat& src, GpuMat& dst, int sp, int sr, TermCriteria criteria, Stream& stream)
{
using namespace ::cv::gpu::cudev::imgproc;
if( src.empty() )
CV_Error( cv::Error::StsBadArg, "The input image is empty" );
if( src.depth() != CV_8U || src.channels() != 4 )
CV_Error( cv::Error::StsUnsupportedFormat, "Only 8-bit, 4-channel images are supported" );
dst.create( src.size(), CV_8UC4 );
if( !(criteria.type & TermCriteria::MAX_ITER) )
criteria.maxCount = 5;
int maxIter = std::min(std::max(criteria.maxCount, 1), 100);
float eps;
if( !(criteria.type & TermCriteria::EPS) )
eps = 1.f;
eps = (float)std::max(criteria.epsilon, 0.0);
meanShiftFiltering_gpu(src, dst, sp, sr, maxIter, eps, StreamAccessor::getStream(stream));
}
////////////////////////////////////////////////////////////////////////
// meanShiftProc_GPU
namespace cv { namespace gpu { namespace cudev
{
namespace imgproc
{
void meanShiftProc_gpu(const PtrStepSzb& src, PtrStepSzb dstr, PtrStepSzb dstsp, int sp, int sr, int maxIter, float eps, cudaStream_t stream);
}
}}}
void cv::gpu::meanShiftProc(const GpuMat& src, GpuMat& dstr, GpuMat& dstsp, int sp, int sr, TermCriteria criteria, Stream& stream)
{
using namespace ::cv::gpu::cudev::imgproc;
if( src.empty() )
CV_Error( cv::Error::StsBadArg, "The input image is empty" );
if( src.depth() != CV_8U || src.channels() != 4 )
CV_Error( cv::Error::StsUnsupportedFormat, "Only 8-bit, 4-channel images are supported" );
dstr.create( src.size(), CV_8UC4 );
dstsp.create( src.size(), CV_16SC2 );
if( !(criteria.type & TermCriteria::MAX_ITER) )
criteria.maxCount = 5;
int maxIter = std::min(std::max(criteria.maxCount, 1), 100);
float eps;
if( !(criteria.type & TermCriteria::EPS) )
eps = 1.f;
eps = (float)std::max(criteria.epsilon, 0.0);
meanShiftProc_gpu(src, dstr, dstsp, sp, sr, maxIter, eps, StreamAccessor::getStream(stream));
}
#endif /* !defined (HAVE_CUDA) */

387
modules/gpuimgproc/src/mssegmentation.cpp
View File

@@ -1,387 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#if !defined HAVE_CUDA || defined(CUDA_DISABLER)
void cv::gpu::meanShiftSegmentation(const GpuMat&, Mat&, int, int, int, TermCriteria) { throw_no_cuda(); }
#else
// Auxiliray stuff
namespace
{
//
// Declarations
//
class DjSets
{
public:
DjSets(int n);
int find(int elem);
int merge(int set1, int set2);
std::vector<int> parent;
std::vector<int> rank;
std::vector<int> size;
private:
DjSets(const DjSets&);
void operator =(const DjSets&);
};
template <typename T>
struct GraphEdge
{
GraphEdge() {}
GraphEdge(int to_, int next_, const T& val_) : to(to_), next(next_), val(val_) {}
int to;
int next;
T val;
};
template <typename T>
class Graph
{
public:
typedef GraphEdge<T> Edge;
Graph(int numv, int nume_max);
void addEdge(int from, int to, const T& val=T());
std::vector<int> start;
std::vector<Edge> edges;
int numv;
int nume_max;
int nume;
private:
Graph(const Graph&);
void operator =(const Graph&);
};
struct SegmLinkVal
{
SegmLinkVal() {}
SegmLinkVal(int dr_, int dsp_) : dr(dr_), dsp(dsp_) {}
bool operator <(const SegmLinkVal& other) const
{
return dr + dsp < other.dr + other.dsp;
}
int dr;
int dsp;
};
struct SegmLink
{
SegmLink() {}
SegmLink(int from_, int to_, const SegmLinkVal& val_)
: from(from_), to(to_), val(val_) {}
bool operator <(const SegmLink& other) const
{
return val < other.val;
}
int from;
int to;
SegmLinkVal val;
};
//
// Implementation
//
DjSets::DjSets(int n) : parent(n), rank(n, 0), size(n, 1)
{
for (int i = 0; i < n; ++i)
parent[i] = i;
}
inline int DjSets::find(int elem)
{
int set = elem;
while (set != parent[set])
set = parent[set];
while (elem != parent[elem])
{
int next = parent[elem];
parent[elem] = set;
elem = next;
}
return set;
}
inline int DjSets::merge(int set1, int set2)
{
if (rank[set1] < rank[set2])
{
parent[set1] = set2;
size[set2] += size[set1];
return set2;
}
if (rank[set2] < rank[set1])
{
parent[set2] = set1;
size[set1] += size[set2];
return set1;
}
parent[set1] = set2;
rank[set2]++;
size[set2] += size[set1];
return set2;
}
template <typename T>
Graph<T>::Graph(int numv_, int nume_max_) : start(numv_, -1), edges(nume_max_)
{
this->numv = numv_;
this->nume_max = nume_max_;
nume = 0;
}
template <typename T>
inline void Graph<T>::addEdge(int from, int to, const T& val)
{
edges[nume] = Edge(to, start[from], val);
start[from] = nume;
nume++;
}
inline int pix(int y, int x, int ncols)
{
return y * ncols + x;
}
inline int sqr(int x)
{
return x * x;
}
inline int dist2(const cv::Vec4b& lhs, const cv::Vec4b& rhs)
{
return sqr(lhs[0] - rhs[0]) + sqr(lhs[1] - rhs[1]) + sqr(lhs[2] - rhs[2]);
}
inline int dist2(const cv::Vec2s& lhs, const cv::Vec2s& rhs)
{
return sqr(lhs[0] - rhs[0]) + sqr(lhs[1] - rhs[1]);
}
} // anonymous namespace
void cv::gpu::meanShiftSegmentation(const GpuMat& src, Mat& dst, int sp, int sr, int minsize, TermCriteria criteria)
{
CV_Assert(src.type() == CV_8UC4);
const int nrows = src.rows;
const int ncols = src.cols;
const int hr = sr;
const int hsp = sp;
// Perform mean shift procedure and obtain region and spatial maps
GpuMat d_rmap, d_spmap;
meanShiftProc(src, d_rmap, d_spmap, sp, sr, criteria);
Mat rmap(d_rmap);
Mat spmap(d_spmap);
Graph<SegmLinkVal> g(nrows * ncols, 4 * (nrows - 1) * (ncols - 1)
+ (nrows - 1) + (ncols - 1));
// Make region adjacent graph from image
Vec4b r1;
Vec4b r2[4];
Vec2s sp1;
Vec2s sp2[4];
int dr[4];
int dsp[4];
for (int y = 0; y < nrows - 1; ++y)
{
Vec4b* ry = rmap.ptr<Vec4b>(y);
Vec4b* ryp = rmap.ptr<Vec4b>(y + 1);
Vec2s* spy = spmap.ptr<Vec2s>(y);
Vec2s* spyp = spmap.ptr<Vec2s>(y + 1);
for (int x = 0; x < ncols - 1; ++x)
{
r1 = ry[x];
sp1 = spy[x];
r2[0] = ry[x + 1];
r2[1] = ryp[x];
r2[2] = ryp[x + 1];
r2[3] = ryp[x];
sp2[0] = spy[x + 1];
sp2[1] = spyp[x];
sp2[2] = spyp[x + 1];
sp2[3] = spyp[x];
dr[0] = dist2(r1, r2[0]);
dr[1] = dist2(r1, r2[1]);
dr[2] = dist2(r1, r2[2]);
dsp[0] = dist2(sp1, sp2[0]);
dsp[1] = dist2(sp1, sp2[1]);
dsp[2] = dist2(sp1, sp2[2]);
r1 = ry[x + 1];
sp1 = spy[x + 1];
dr[3] = dist2(r1, r2[3]);
dsp[3] = dist2(sp1, sp2[3]);
g.addEdge(pix(y, x, ncols), pix(y, x + 1, ncols), SegmLinkVal(dr[0], dsp[0]));
g.addEdge(pix(y, x, ncols), pix(y + 1, x, ncols), SegmLinkVal(dr[1], dsp[1]));
g.addEdge(pix(y, x, ncols), pix(y + 1, x + 1, ncols), SegmLinkVal(dr[2], dsp[2]));
g.addEdge(pix(y, x + 1, ncols), pix(y + 1, x, ncols), SegmLinkVal(dr[3], dsp[3]));
}
}
for (int y = 0; y < nrows - 1; ++y)
{
r1 = rmap.at<Vec4b>(y, ncols - 1);
r2[0] = rmap.at<Vec4b>(y + 1, ncols - 1);
sp1 = spmap.at<Vec2s>(y, ncols - 1);
sp2[0] = spmap.at<Vec2s>(y + 1, ncols - 1);
dr[0] = dist2(r1, r2[0]);
dsp[0] = dist2(sp1, sp2[0]);
g.addEdge(pix(y, ncols - 1, ncols), pix(y + 1, ncols - 1, ncols), SegmLinkVal(dr[0], dsp[0]));
}
for (int x = 0; x < ncols - 1; ++x)
{
r1 = rmap.at<Vec4b>(nrows - 1, x);
r2[0] = rmap.at<Vec4b>(nrows - 1, x + 1);
sp1 = spmap.at<Vec2s>(nrows - 1, x);
sp2[0] = spmap.at<Vec2s>(nrows - 1, x + 1);
dr[0] = dist2(r1, r2[0]);
dsp[0] = dist2(sp1, sp2[0]);
g.addEdge(pix(nrows - 1, x, ncols), pix(nrows - 1, x + 1, ncols), SegmLinkVal(dr[0], dsp[0]));
}
DjSets comps(g.numv);
// Find adjacent components
for (int v = 0; v < g.numv; ++v)
{
for (int e_it = g.start[v]; e_it != -1; e_it = g.edges[e_it].next)
{
int c1 = comps.find(v);
int c2 = comps.find(g.edges[e_it].to);
if (c1 != c2 && g.edges[e_it].val.dr < hr && g.edges[e_it].val.dsp < hsp)
comps.merge(c1, c2);
}
}
std::vector<SegmLink> edges;
edges.reserve(g.numv);
// Prepare edges connecting differnet components
for (int v = 0; v < g.numv; ++v)
{
int c1 = comps.find(v);
for (int e_it = g.start[v]; e_it != -1; e_it = g.edges[e_it].next)
{
int c2 = comps.find(g.edges[e_it].to);
if (c1 != c2)
edges.push_back(SegmLink(c1, c2, g.edges[e_it].val));
}
}
// Sort all graph's edges connecting differnet components (in asceding order)
sort(edges.begin(), edges.end());
// Exclude small components (starting from the nearest couple)
for (size_t i = 0; i < edges.size(); ++i)
{
int c1 = comps.find(edges[i].from);
int c2 = comps.find(edges[i].to);
if (c1 != c2 && (comps.size[c1] < minsize || comps.size[c2] < minsize))
comps.merge(c1, c2);
}
// Compute sum of the pixel's colors which are in the same segment
Mat h_src(src);
std::vector<Vec4i> sumcols(nrows * ncols, Vec4i(0, 0, 0, 0));
for (int y = 0; y < nrows; ++y)
{
Vec4b* h_srcy = h_src.ptr<Vec4b>(y);
for (int x = 0; x < ncols; ++x)
{
int parent = comps.find(pix(y, x, ncols));
Vec4b col = h_srcy[x];
Vec4i& sumcol = sumcols[parent];
sumcol[0] += col[0];
sumcol[1] += col[1];
sumcol[2] += col[2];
}
}
// Create final image, color of each segment is the average color of its pixels
dst.create(src.size(), src.type());
for (int y = 0; y < nrows; ++y)
{
Vec4b* dsty = dst.ptr<Vec4b>(y);
for (int x = 0; x < ncols; ++x)
{
int parent = comps.find(pix(y, x, ncols));
const Vec4i& sumcol = sumcols[parent];
Vec4b& dstcol = dsty[x];
dstcol[0] = static_cast<uchar>(sumcol[0] / comps.size[parent]);
dstcol[1] = static_cast<uchar>(sumcol[1] / comps.size[parent]);
dstcol[2] = static_cast<uchar>(sumcol[2] / comps.size[parent]);
dstcol[3] = 255;
}
}
}
#endif // #if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)

43
modules/gpuimgproc/src/precomp.cpp
View File

@@ -1,43 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"

58
modules/gpuimgproc/src/precomp.hpp
View File

@@ -1,58 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_PRECOMP_H__
#define __OPENCV_PRECOMP_H__
#include "opencv2/gpuimgproc.hpp"
#include "opencv2/gpufilters.hpp"
#include "opencv2/core/private.hpp"
#include "opencv2/core/gpu_private.hpp"
#include "opencv2/opencv_modules.hpp"
#ifdef HAVE_OPENCV_GPUARITHM
# include "opencv2/gpuarithm.hpp"
#endif
#endif /* __OPENCV_PRECOMP_H__ */

97
modules/gpuimgproc/test/test_bilateral_filter.cpp
View File

@@ -1,97 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
////////////////////////////////////////////////////////
// BilateralFilter
PARAM_TEST_CASE(BilateralFilter, cv::gpu::DeviceInfo, cv::Size, MatType)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
int type;
int kernel_size;
float sigma_color;
float sigma_spatial;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
type = GET_PARAM(2);
kernel_size = 5;
sigma_color = 10.f;
sigma_spatial = 3.5f;
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(BilateralFilter, Accuracy)
{
cv::Mat src = randomMat(size, type);
src.convertTo(src, type);
cv::gpu::GpuMat dst;
cv::gpu::bilateralFilter(loadMat(src), dst, kernel_size, sigma_color, sigma_spatial);
cv::Mat dst_gold;
cv::bilateralFilter(src, dst_gold, kernel_size, sigma_color, sigma_spatial);
EXPECT_MAT_NEAR(dst_gold, dst, src.depth() == CV_32F ? 1e-3 : 1.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, BilateralFilter, testing::Combine(
ALL_DEVICES,
testing::Values(cv::Size(128, 128), cv::Size(113, 113), cv::Size(639, 481)),
testing::Values(MatType(CV_8UC1), MatType(CV_8UC3), MatType(CV_32FC1), MatType(CV_32FC3))
));
#endif // HAVE_CUDA

124
modules/gpuimgproc/test/test_blend.cpp
View File

@@ -1,124 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
////////////////////////////////////////////////////////////////////////////
// Blend
namespace
{
template <typename T>
void blendLinearGold(const cv::Mat& img1, const cv::Mat& img2, const cv::Mat& weights1, const cv::Mat& weights2, cv::Mat& result_gold)
{
result_gold.create(img1.size(), img1.type());
int cn = img1.channels();
for (int y = 0; y < img1.rows; ++y)
{
const float* weights1_row = weights1.ptr<float>(y);
const float* weights2_row = weights2.ptr<float>(y);
const T* img1_row = img1.ptr<T>(y);
const T* img2_row = img2.ptr<T>(y);
T* result_gold_row = result_gold.ptr<T>(y);
for (int x = 0; x < img1.cols * cn; ++x)
{
float w1 = weights1_row[x / cn];
float w2 = weights2_row[x / cn];
result_gold_row[x] = static_cast<T>((img1_row[x] * w1 + img2_row[x] * w2) / (w1 + w2 + 1e-5f));
}
}
}
}
PARAM_TEST_CASE(Blend, cv::gpu::DeviceInfo, cv::Size, MatType, UseRoi)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
int type;
bool useRoi;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
type = GET_PARAM(2);
useRoi = GET_PARAM(3);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(Blend, Accuracy)
{
int depth = CV_MAT_DEPTH(type);
cv::Mat img1 = randomMat(size, type, 0.0, depth == CV_8U ? 255.0 : 1.0);
cv::Mat img2 = randomMat(size, type, 0.0, depth == CV_8U ? 255.0 : 1.0);
cv::Mat weights1 = randomMat(size, CV_32F, 0, 1);
cv::Mat weights2 = randomMat(size, CV_32F, 0, 1);
cv::gpu::GpuMat result;
cv::gpu::blendLinear(loadMat(img1, useRoi), loadMat(img2, useRoi), loadMat(weights1, useRoi), loadMat(weights2, useRoi), result);
cv::Mat result_gold;
if (depth == CV_8U)
blendLinearGold<uchar>(img1, img2, weights1, weights2, result_gold);
else
blendLinearGold<float>(img1, img2, weights1, weights2, result_gold);
EXPECT_MAT_NEAR(result_gold, result, CV_MAT_DEPTH(type) == CV_8U ? 1.0 : 1e-5);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Blend, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
testing::Values(MatType(CV_8UC1), MatType(CV_8UC3), MatType(CV_8UC4), MatType(CV_32FC1), MatType(CV_32FC3), MatType(CV_32FC4)),
WHOLE_SUBMAT));
#endif // HAVE_CUDA

114
modules/gpuimgproc/test/test_canny.cpp
View File

@@ -1,114 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
////////////////////////////////////////////////////////
// Canny
namespace
{
IMPLEMENT_PARAM_CLASS(AppertureSize, int)
IMPLEMENT_PARAM_CLASS(L2gradient, bool)
}
PARAM_TEST_CASE(Canny, cv::gpu::DeviceInfo, AppertureSize, L2gradient, UseRoi)
{
cv::gpu::DeviceInfo devInfo;
int apperture_size;
bool useL2gradient;
bool useRoi;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
apperture_size = GET_PARAM(1);
useL2gradient = GET_PARAM(2);
useRoi = GET_PARAM(3);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(Canny, Accuracy)
{
cv::Mat img = readImage("stereobm/aloe-L.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
double low_thresh = 50.0;
double high_thresh = 100.0;
if (!supportFeature(devInfo, cv::gpu::SHARED_ATOMICS))
{
try
{
cv::gpu::GpuMat edges;
cv::gpu::Canny(loadMat(img), edges, low_thresh, high_thresh, apperture_size, useL2gradient);
}
catch (const cv::Exception& e)
{
ASSERT_EQ(cv::Error::StsNotImplemented, e.code);
}
}
else
{
cv::gpu::GpuMat edges;
cv::gpu::Canny(loadMat(img, useRoi), edges, low_thresh, high_thresh, apperture_size, useL2gradient);
cv::Mat edges_gold;
cv::Canny(img, edges_gold, low_thresh, high_thresh, apperture_size, useL2gradient);
EXPECT_MAT_SIMILAR(edges_gold, edges, 2e-2);
}
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Canny, testing::Combine(
ALL_DEVICES,
testing::Values(AppertureSize(3), AppertureSize(5)),
testing::Values(L2gradient(false), L2gradient(true)),
WHOLE_SUBMAT));
#endif // HAVE_CUDA

2503
modules/gpuimgproc/test/test_color.cpp
View File

File diff suppressed because it is too large Load Diff

145
modules/gpuimgproc/test/test_corners.cpp
View File

@@ -1,145 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
///////////////////////////////////////////////////////////////////////////////////////////////////////
// CornerHarris
namespace
{
IMPLEMENT_PARAM_CLASS(BlockSize, int);
IMPLEMENT_PARAM_CLASS(ApertureSize, int);
}
PARAM_TEST_CASE(CornerHarris, cv::gpu::DeviceInfo, MatType, BorderType, BlockSize, ApertureSize)
{
cv::gpu::DeviceInfo devInfo;
int type;
int borderType;
int blockSize;
int apertureSize;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
type = GET_PARAM(1);
borderType = GET_PARAM(2);
blockSize = GET_PARAM(3);
apertureSize = GET_PARAM(4);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(CornerHarris, Accuracy)
{
cv::Mat src = readImageType("stereobm/aloe-L.png", type);
ASSERT_FALSE(src.empty());
double k = randomDouble(0.1, 0.9);
cv::gpu::GpuMat dst;
cv::gpu::cornerHarris(loadMat(src), dst, blockSize, apertureSize, k, borderType);
cv::Mat dst_gold;
cv::cornerHarris(src, dst_gold, blockSize, apertureSize, k, borderType);
EXPECT_MAT_NEAR(dst_gold, dst, 0.02);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CornerHarris, testing::Combine(
ALL_DEVICES,
testing::Values(MatType(CV_8UC1), MatType(CV_32FC1)),
testing::Values(BorderType(cv::BORDER_REFLECT101), BorderType(cv::BORDER_REPLICATE), BorderType(cv::BORDER_REFLECT)),
testing::Values(BlockSize(3), BlockSize(5), BlockSize(7)),
testing::Values(ApertureSize(0), ApertureSize(3), ApertureSize(5), ApertureSize(7))));
///////////////////////////////////////////////////////////////////////////////////////////////////////
// cornerMinEigen
PARAM_TEST_CASE(CornerMinEigen, cv::gpu::DeviceInfo, MatType, BorderType, BlockSize, ApertureSize)
{
cv::gpu::DeviceInfo devInfo;
int type;
int borderType;
int blockSize;
int apertureSize;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
type = GET_PARAM(1);
borderType = GET_PARAM(2);
blockSize = GET_PARAM(3);
apertureSize = GET_PARAM(4);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(CornerMinEigen, Accuracy)
{
cv::Mat src = readImageType("stereobm/aloe-L.png", type);
ASSERT_FALSE(src.empty());
cv::gpu::GpuMat dst;
cv::gpu::cornerMinEigenVal(loadMat(src), dst, blockSize, apertureSize, borderType);
cv::Mat dst_gold;
cv::cornerMinEigenVal(src, dst_gold, blockSize, apertureSize, borderType);
EXPECT_MAT_NEAR(dst_gold, dst, 0.02);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CornerMinEigen, testing::Combine(
ALL_DEVICES,
testing::Values(MatType(CV_8UC1), MatType(CV_32FC1)),
testing::Values(BorderType(cv::BORDER_REFLECT101), BorderType(cv::BORDER_REPLICATE), BorderType(cv::BORDER_REFLECT)),
testing::Values(BlockSize(3), BlockSize(5), BlockSize(7)),
testing::Values(ApertureSize(0), ApertureSize(3), ApertureSize(5), ApertureSize(7))));
#endif // HAVE_CUDA

131
modules/gpuimgproc/test/test_gftt.cpp
View File

@@ -1,131 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
//////////////////////////////////////////////////////
// GoodFeaturesToTrack
namespace
{
IMPLEMENT_PARAM_CLASS(MinDistance, double)
}
PARAM_TEST_CASE(GoodFeaturesToTrack, cv::gpu::DeviceInfo, MinDistance)
{
cv::gpu::DeviceInfo devInfo;
double minDistance;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
minDistance = GET_PARAM(1);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(GoodFeaturesToTrack, Accuracy)
{
cv::Mat image = readImage("opticalflow/frame0.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
int maxCorners = 1000;
double qualityLevel = 0.01;
cv::gpu::GoodFeaturesToTrackDetector_GPU detector(maxCorners, qualityLevel, minDistance);
cv::gpu::GpuMat d_pts;
detector(loadMat(image), d_pts);
ASSERT_FALSE(d_pts.empty());
std::vector<cv::Point2f> pts(d_pts.cols);
cv::Mat pts_mat(1, d_pts.cols, CV_32FC2, (void*) &pts[0]);
d_pts.download(pts_mat);
std::vector<cv::Point2f> pts_gold;
cv::goodFeaturesToTrack(image, pts_gold, maxCorners, qualityLevel, minDistance);
ASSERT_EQ(pts_gold.size(), pts.size());
size_t mistmatch = 0;
for (size_t i = 0; i < pts.size(); ++i)
{
cv::Point2i a = pts_gold[i];
cv::Point2i b = pts[i];
bool eq = std::abs(a.x - b.x) < 1 && std::abs(a.y - b.y) < 1;
if (!eq)
++mistmatch;
}
double bad_ratio = static_cast<double>(mistmatch) / pts.size();
ASSERT_LE(bad_ratio, 0.01);
}
GPU_TEST_P(GoodFeaturesToTrack, EmptyCorners)
{
int maxCorners = 1000;
double qualityLevel = 0.01;
cv::gpu::GoodFeaturesToTrackDetector_GPU detector(maxCorners, qualityLevel, minDistance);
cv::gpu::GpuMat src(100, 100, CV_8UC1, cv::Scalar::all(0));
cv::gpu::GpuMat corners(1, maxCorners, CV_32FC2);
detector(src, corners);
ASSERT_TRUE(corners.empty());
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, GoodFeaturesToTrack, testing::Combine(
ALL_DEVICES,
testing::Values(MinDistance(0.0), MinDistance(3.0))));
#endif // HAVE_CUDA

227
modules/gpuimgproc/test/test_histogram.cpp
View File

@@ -1,227 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
///////////////////////////////////////////////////////////////////////////////////////////////////////
// HistEven
struct HistEven : testing::TestWithParam<cv::gpu::DeviceInfo>
{
cv::gpu::DeviceInfo devInfo;
virtual void SetUp()
{
devInfo = GetParam();
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(HistEven, Accuracy)
{
cv::Mat img = readImage("stereobm/aloe-L.png");
ASSERT_FALSE(img.empty());
cv::Mat hsv;
cv::cvtColor(img, hsv, cv::COLOR_BGR2HSV);
int hbins = 30;
float hranges[] = {0.0f, 180.0f};
std::vector<cv::gpu::GpuMat> srcs;
cv::gpu::split(loadMat(hsv), srcs);
cv::gpu::GpuMat hist;
cv::gpu::histEven(srcs[0], hist, hbins, (int)hranges[0], (int)hranges[1]);
cv::MatND histnd;
int histSize[] = {hbins};
const float* ranges[] = {hranges};
int channels[] = {0};
cv::calcHist(&hsv, 1, channels, cv::Mat(), histnd, 1, histSize, ranges);
cv::Mat hist_gold = histnd;
hist_gold = hist_gold.t();
hist_gold.convertTo(hist_gold, CV_32S);
EXPECT_MAT_NEAR(hist_gold, hist, 0.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, HistEven, ALL_DEVICES);
///////////////////////////////////////////////////////////////////////////////////////////////////////
// CalcHist
namespace
{
void calcHistGold(const cv::Mat& src, cv::Mat& hist)
{
hist.create(1, 256, CV_32SC1);
hist.setTo(cv::Scalar::all(0));
int* hist_row = hist.ptr<int>();
for (int y = 0; y < src.rows; ++y)
{
const uchar* src_row = src.ptr(y);
for (int x = 0; x < src.cols; ++x)
++hist_row[src_row[x]];
}
}
}
PARAM_TEST_CASE(CalcHist, cv::gpu::DeviceInfo, cv::Size)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(CalcHist, Accuracy)
{
cv::Mat src = randomMat(size, CV_8UC1);
cv::gpu::GpuMat hist;
cv::gpu::calcHist(loadMat(src), hist);
cv::Mat hist_gold;
calcHistGold(src, hist_gold);
EXPECT_MAT_NEAR(hist_gold, hist, 0.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CalcHist, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES));
///////////////////////////////////////////////////////////////////////////////////////////////////////
// EqualizeHist
PARAM_TEST_CASE(EqualizeHist, cv::gpu::DeviceInfo, cv::Size)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(EqualizeHist, Accuracy)
{
cv::Mat src = randomMat(size, CV_8UC1);
cv::gpu::GpuMat dst;
cv::gpu::equalizeHist(loadMat(src), dst);
cv::Mat dst_gold;
cv::equalizeHist(src, dst_gold);
EXPECT_MAT_NEAR(dst_gold, dst, 3.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, EqualizeHist, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES));
///////////////////////////////////////////////////////////////////////////////////////////////////////
// CLAHE
namespace
{
IMPLEMENT_PARAM_CLASS(ClipLimit, double)
}
PARAM_TEST_CASE(CLAHE, cv::gpu::DeviceInfo, cv::Size, ClipLimit)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
double clipLimit;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
clipLimit = GET_PARAM(2);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(CLAHE, Accuracy)
{
cv::Mat src = randomMat(size, CV_8UC1);
cv::Ptr<cv::gpu::CLAHE> clahe = cv::gpu::createCLAHE(clipLimit);
cv::gpu::GpuMat dst;
clahe->apply(loadMat(src), dst);
cv::Ptr<cv::CLAHE> clahe_gold = cv::createCLAHE(clipLimit);
cv::Mat dst_gold;
clahe_gold->apply(src, dst_gold);
ASSERT_MAT_NEAR(dst_gold, dst, 1.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CLAHE, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
testing::Values(0.0, 40.0)));
#endif // HAVE_CUDA

255
modules/gpuimgproc/test/test_hough.cpp
View File

@@ -1,255 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
///////////////////////////////////////////////////////////////////////////////////////////////////////
// HoughLines
PARAM_TEST_CASE(HoughLines, cv::gpu::DeviceInfo, cv::Size, UseRoi)
{
static void generateLines(cv::Mat& img)
{
img.setTo(cv::Scalar::all(0));
cv::line(img, cv::Point(20, 0), cv::Point(20, img.rows), cv::Scalar::all(255));
cv::line(img, cv::Point(0, 50), cv::Point(img.cols, 50), cv::Scalar::all(255));
cv::line(img, cv::Point(0, 0), cv::Point(img.cols, img.rows), cv::Scalar::all(255));
cv::line(img, cv::Point(img.cols, 0), cv::Point(0, img.rows), cv::Scalar::all(255));
}
static void drawLines(cv::Mat& dst, const std::vector<cv::Vec2f>& lines)
{
dst.setTo(cv::Scalar::all(0));
for (size_t i = 0; i < lines.size(); ++i)
{
float rho = lines[i][0], theta = lines[i][1];
cv::Point pt1, pt2;
double a = std::cos(theta), b = std::sin(theta);
double x0 = a*rho, y0 = b*rho;
pt1.x = cvRound(x0 + 1000*(-b));
pt1.y = cvRound(y0 + 1000*(a));
pt2.x = cvRound(x0 - 1000*(-b));
pt2.y = cvRound(y0 - 1000*(a));
cv::line(dst, pt1, pt2, cv::Scalar::all(255));
}
}
};
GPU_TEST_P(HoughLines, Accuracy)
{
const cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
cv::gpu::setDevice(devInfo.deviceID());
const cv::Size size = GET_PARAM(1);
const bool useRoi = GET_PARAM(2);
const float rho = 1.0f;
const float theta = (float) (1.5 * CV_PI / 180.0);
const int threshold = 100;
cv::Mat src(size, CV_8UC1);
generateLines(src);
cv::gpu::GpuMat d_lines;
cv::gpu::HoughLines(loadMat(src, useRoi), d_lines, rho, theta, threshold);
std::vector<cv::Vec2f> lines;
cv::gpu::HoughLinesDownload(d_lines, lines);
cv::Mat dst(size, CV_8UC1);
drawLines(dst, lines);
ASSERT_MAT_NEAR(src, dst, 0.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, HoughLines, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
WHOLE_SUBMAT));
///////////////////////////////////////////////////////////////////////////////////////////////////////
// HoughCircles
PARAM_TEST_CASE(HoughCircles, cv::gpu::DeviceInfo, cv::Size, UseRoi)
{
static void drawCircles(cv::Mat& dst, const std::vector<cv::Vec3f>& circles, bool fill)
{
dst.setTo(cv::Scalar::all(0));
for (size_t i = 0; i < circles.size(); ++i)
cv::circle(dst, cv::Point2f(circles[i][0], circles[i][1]), (int)circles[i][2], cv::Scalar::all(255), fill ? -1 : 1);
}
};
GPU_TEST_P(HoughCircles, Accuracy)
{
const cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
cv::gpu::setDevice(devInfo.deviceID());
const cv::Size size = GET_PARAM(1);
const bool useRoi = GET_PARAM(2);
const float dp = 2.0f;
const float minDist = 0.0f;
const int minRadius = 10;
const int maxRadius = 20;
const int cannyThreshold = 100;
const int votesThreshold = 20;
std::vector<cv::Vec3f> circles_gold(4);
circles_gold[0] = cv::Vec3i(20, 20, minRadius);
circles_gold[1] = cv::Vec3i(90, 87, minRadius + 3);
circles_gold[2] = cv::Vec3i(30, 70, minRadius + 8);
circles_gold[3] = cv::Vec3i(80, 10, maxRadius);
cv::Mat src(size, CV_8UC1);
drawCircles(src, circles_gold, true);
cv::gpu::GpuMat d_circles;
cv::gpu::HoughCircles(loadMat(src, useRoi), d_circles, cv::HOUGH_GRADIENT, dp, minDist, cannyThreshold, votesThreshold, minRadius, maxRadius);
std::vector<cv::Vec3f> circles;
cv::gpu::HoughCirclesDownload(d_circles, circles);
ASSERT_FALSE(circles.empty());
for (size_t i = 0; i < circles.size(); ++i)
{
cv::Vec3f cur = circles[i];
bool found = false;
for (size_t j = 0; j < circles_gold.size(); ++j)
{
cv::Vec3f gold = circles_gold[j];
if (std::fabs(cur[0] - gold[0]) < 5 && std::fabs(cur[1] - gold[1]) < 5 && std::fabs(cur[2] - gold[2]) < 5)
{
found = true;
break;
}
}
ASSERT_TRUE(found);
}
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, HoughCircles, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
WHOLE_SUBMAT));
///////////////////////////////////////////////////////////////////////////////////////////////////////
// GeneralizedHough
PARAM_TEST_CASE(GeneralizedHough, cv::gpu::DeviceInfo, UseRoi)
{
};
GPU_TEST_P(GeneralizedHough, POSITION)
{
const cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
cv::gpu::setDevice(devInfo.deviceID());
const bool useRoi = GET_PARAM(1);
cv::Mat templ = readImage("../cv/shared/templ.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(templ.empty());
cv::Point templCenter(templ.cols / 2, templ.rows / 2);
const size_t gold_count = 3;
cv::Point pos_gold[gold_count];
pos_gold[0] = cv::Point(templCenter.x + 10, templCenter.y + 10);
pos_gold[1] = cv::Point(2 * templCenter.x + 40, templCenter.y + 10);
pos_gold[2] = cv::Point(2 * templCenter.x + 40, 2 * templCenter.y + 40);
cv::Mat image(templ.rows * 3, templ.cols * 3, CV_8UC1, cv::Scalar::all(0));
for (size_t i = 0; i < gold_count; ++i)
{
cv::Rect rec(pos_gold[i].x - templCenter.x, pos_gold[i].y - templCenter.y, templ.cols, templ.rows);
cv::Mat imageROI = image(rec);
templ.copyTo(imageROI);
}
cv::Ptr<cv::gpu::GeneralizedHough_GPU> hough = cv::gpu::GeneralizedHough_GPU::create(cv::GeneralizedHough::GHT_POSITION);
hough->set("votesThreshold", 200);
hough->setTemplate(loadMat(templ, useRoi));
cv::gpu::GpuMat d_pos;
hough->detect(loadMat(image, useRoi), d_pos);
std::vector<cv::Vec4f> pos;
hough->download(d_pos, pos);
ASSERT_EQ(gold_count, pos.size());
for (size_t i = 0; i < gold_count; ++i)
{
cv::Point gold = pos_gold[i];
bool found = false;
for (size_t j = 0; j < pos.size(); ++j)
{
cv::Point2f p(pos[j][0], pos[j][1]);
if (::fabs(p.x - gold.x) < 2 && ::fabs(p.y - gold.y) < 2)
{
found = true;
break;
}
}
ASSERT_TRUE(found);
}
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, GeneralizedHough, testing::Combine(
ALL_DEVICES,
WHOLE_SUBMAT));
#endif // HAVE_CUDA

45
modules/gpuimgproc/test/test_main.cpp
View File

@@ -1,45 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
CV_GPU_TEST_MAIN("gpu")

305
modules/gpuimgproc/test/test_match_template.cpp
View File

@@ -1,305 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate8U
CV_ENUM(TemplateMethod, cv::TM_SQDIFF, cv::TM_SQDIFF_NORMED, cv::TM_CCORR, cv::TM_CCORR_NORMED, cv::TM_CCOEFF, cv::TM_CCOEFF_NORMED)
#define ALL_TEMPLATE_METHODS testing::Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_SQDIFF_NORMED), TemplateMethod(cv::TM_CCORR), TemplateMethod(cv::TM_CCORR_NORMED), TemplateMethod(cv::TM_CCOEFF), TemplateMethod(cv::TM_CCOEFF_NORMED))
namespace
{
IMPLEMENT_PARAM_CLASS(TemplateSize, cv::Size);
}
PARAM_TEST_CASE(MatchTemplate8U, cv::gpu::DeviceInfo, cv::Size, TemplateSize, Channels, TemplateMethod)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
cv::Size templ_size;
int cn;
int method;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
templ_size = GET_PARAM(2);
cn = GET_PARAM(3);
method = GET_PARAM(4);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(MatchTemplate8U, Accuracy)
{
cv::Mat image = randomMat(size, CV_MAKETYPE(CV_8U, cn));
cv::Mat templ = randomMat(templ_size, CV_MAKETYPE(CV_8U, cn));
cv::gpu::GpuMat dst;
cv::gpu::matchTemplate(loadMat(image), loadMat(templ), dst, method);
cv::Mat dst_gold;
cv::matchTemplate(image, templ, dst_gold, method);
EXPECT_MAT_NEAR(dst_gold, dst, templ_size.area() * 1e-1);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate8U, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
testing::Values(TemplateSize(cv::Size(5, 5)), TemplateSize(cv::Size(16, 16)), TemplateSize(cv::Size(30, 30))),
testing::Values(Channels(1), Channels(3), Channels(4)),
ALL_TEMPLATE_METHODS));
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate32F
PARAM_TEST_CASE(MatchTemplate32F, cv::gpu::DeviceInfo, cv::Size, TemplateSize, Channels, TemplateMethod)
{
cv::gpu::DeviceInfo devInfo;
cv::Size size;
cv::Size templ_size;
int cn;
int method;
int n, m, h, w;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
size = GET_PARAM(1);
templ_size = GET_PARAM(2);
cn = GET_PARAM(3);
method = GET_PARAM(4);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(MatchTemplate32F, Regression)
{
cv::Mat image = randomMat(size, CV_MAKETYPE(CV_32F, cn));
cv::Mat templ = randomMat(templ_size, CV_MAKETYPE(CV_32F, cn));
cv::gpu::GpuMat dst;
cv::gpu::matchTemplate(loadMat(image), loadMat(templ), dst, method);
cv::Mat dst_gold;
cv::matchTemplate(image, templ, dst_gold, method);
EXPECT_MAT_NEAR(dst_gold, dst, templ_size.area() * 1e-1);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate32F, testing::Combine(
ALL_DEVICES,
DIFFERENT_SIZES,
testing::Values(TemplateSize(cv::Size(5, 5)), TemplateSize(cv::Size(16, 16)), TemplateSize(cv::Size(30, 30))),
testing::Values(Channels(1), Channels(3), Channels(4)),
testing::Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_CCORR))));
////////////////////////////////////////////////////////////////////////////////
// MatchTemplateBlackSource
PARAM_TEST_CASE(MatchTemplateBlackSource, cv::gpu::DeviceInfo, TemplateMethod)
{
cv::gpu::DeviceInfo devInfo;
int method;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
method = GET_PARAM(1);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(MatchTemplateBlackSource, Accuracy)
{
cv::Mat image = readImage("matchtemplate/black.png");
ASSERT_FALSE(image.empty());
cv::Mat pattern = readImage("matchtemplate/cat.png");
ASSERT_FALSE(pattern.empty());
cv::gpu::GpuMat d_dst;
cv::gpu::matchTemplate(loadMat(image), loadMat(pattern), d_dst, method);
cv::Mat dst(d_dst);
double maxValue;
cv::Point maxLoc;
cv::minMaxLoc(dst, NULL, &maxValue, NULL, &maxLoc);
cv::Point maxLocGold = cv::Point(284, 12);
ASSERT_EQ(maxLocGold, maxLoc);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplateBlackSource, testing::Combine(
ALL_DEVICES,
testing::Values(TemplateMethod(cv::TM_CCOEFF_NORMED), TemplateMethod(cv::TM_CCORR_NORMED))));
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate_CCOEF_NORMED
PARAM_TEST_CASE(MatchTemplate_CCOEF_NORMED, cv::gpu::DeviceInfo, std::pair<std::string, std::string>)
{
cv::gpu::DeviceInfo devInfo;
std::string imageName;
std::string patternName;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
imageName = GET_PARAM(1).first;
patternName = GET_PARAM(1).second;
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(MatchTemplate_CCOEF_NORMED, Accuracy)
{
cv::Mat image = readImage(imageName);
ASSERT_FALSE(image.empty());
cv::Mat pattern = readImage(patternName);
ASSERT_FALSE(pattern.empty());
cv::gpu::GpuMat d_dst;
cv::gpu::matchTemplate(loadMat(image), loadMat(pattern), d_dst, cv::TM_CCOEFF_NORMED);
cv::Mat dst(d_dst);
cv::Point minLoc, maxLoc;
double minVal, maxVal;
cv::minMaxLoc(dst, &minVal, &maxVal, &minLoc, &maxLoc);
cv::Mat dstGold;
cv::matchTemplate(image, pattern, dstGold, cv::TM_CCOEFF_NORMED);
double minValGold, maxValGold;
cv::Point minLocGold, maxLocGold;
cv::minMaxLoc(dstGold, &minValGold, &maxValGold, &minLocGold, &maxLocGold);
ASSERT_EQ(minLocGold, minLoc);
ASSERT_EQ(maxLocGold, maxLoc);
ASSERT_LE(maxVal, 1.0);
ASSERT_GE(minVal, -1.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate_CCOEF_NORMED, testing::Combine(
ALL_DEVICES,
testing::Values(std::make_pair(std::string("matchtemplate/source-0.png"), std::string("matchtemplate/target-0.png")))));
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate_CanFindBigTemplate
struct MatchTemplate_CanFindBigTemplate : testing::TestWithParam<cv::gpu::DeviceInfo>
{
cv::gpu::DeviceInfo devInfo;
virtual void SetUp()
{
devInfo = GetParam();
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(MatchTemplate_CanFindBigTemplate, SQDIFF_NORMED)
{
cv::Mat scene = readImage("matchtemplate/scene.png");
ASSERT_FALSE(scene.empty());
cv::Mat templ = readImage("matchtemplate/template.png");
ASSERT_FALSE(templ.empty());
cv::gpu::GpuMat d_result;
cv::gpu::matchTemplate(loadMat(scene), loadMat(templ), d_result, cv::TM_SQDIFF_NORMED);
cv::Mat result(d_result);
double minVal;
cv::Point minLoc;
cv::minMaxLoc(result, &minVal, 0, &minLoc, 0);
ASSERT_GE(minVal, 0);
ASSERT_LT(minVal, 1e-3);
ASSERT_EQ(344, minLoc.x);
ASSERT_EQ(0, minLoc.y);
}
GPU_TEST_P(MatchTemplate_CanFindBigTemplate, SQDIFF)
{
cv::Mat scene = readImage("matchtemplate/scene.png");
ASSERT_FALSE(scene.empty());
cv::Mat templ = readImage("matchtemplate/template.png");
ASSERT_FALSE(templ.empty());
cv::gpu::GpuMat d_result;
cv::gpu::matchTemplate(loadMat(scene), loadMat(templ), d_result, cv::TM_SQDIFF);
cv::Mat result(d_result);
double minVal;
cv::Point minLoc;
cv::minMaxLoc(result, &minVal, 0, &minLoc, 0);
ASSERT_GE(minVal, 0);
ASSERT_EQ(344, minLoc.x);
ASSERT_EQ(0, minLoc.y);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate_CanFindBigTemplate, ALL_DEVICES);
#endif // HAVE_CUDA

174
modules/gpuimgproc/test/test_mean_shift.cpp
View File

@@ -1,174 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace cvtest;
////////////////////////////////////////////////////////////////////////////////
// MeanShift
struct MeanShift : testing::TestWithParam<cv::gpu::DeviceInfo>
{
cv::gpu::DeviceInfo devInfo;
cv::Mat img;
int spatialRad;
int colorRad;
virtual void SetUp()
{
devInfo = GetParam();
cv::gpu::setDevice(devInfo.deviceID());
img = readImageType("meanshift/cones.png", CV_8UC4);
ASSERT_FALSE(img.empty());
spatialRad = 30;
colorRad = 30;
}
};
GPU_TEST_P(MeanShift, Filtering)
{
cv::Mat img_template;
if (supportFeature(devInfo, cv::gpu::FEATURE_SET_COMPUTE_20))
img_template = readImage("meanshift/con_result.png");
else
img_template = readImage("meanshift/con_result_CC1X.png");
ASSERT_FALSE(img_template.empty());
cv::gpu::GpuMat d_dst;
cv::gpu::meanShiftFiltering(loadMat(img), d_dst, spatialRad, colorRad);
ASSERT_EQ(CV_8UC4, d_dst.type());
cv::Mat dst(d_dst);
cv::Mat result;
cv::cvtColor(dst, result, cv::COLOR_BGRA2BGR);
EXPECT_MAT_NEAR(img_template, result, 0.0);
}
GPU_TEST_P(MeanShift, Proc)
{
cv::FileStorage fs;
if (supportFeature(devInfo, cv::gpu::FEATURE_SET_COMPUTE_20))
fs.open(std::string(cvtest::TS::ptr()->get_data_path()) + "meanshift/spmap.yaml", cv::FileStorage::READ);
else
fs.open(std::string(cvtest::TS::ptr()->get_data_path()) + "meanshift/spmap_CC1X.yaml", cv::FileStorage::READ);
ASSERT_TRUE(fs.isOpened());
cv::Mat spmap_template;
fs["spmap"] >> spmap_template;
ASSERT_FALSE(spmap_template.empty());
cv::gpu::GpuMat rmap_filtered;
cv::gpu::meanShiftFiltering(loadMat(img), rmap_filtered, spatialRad, colorRad);
cv::gpu::GpuMat rmap;
cv::gpu::GpuMat spmap;
cv::gpu::meanShiftProc(loadMat(img), rmap, spmap, spatialRad, colorRad);
ASSERT_EQ(CV_8UC4, rmap.type());
EXPECT_MAT_NEAR(rmap_filtered, rmap, 0.0);
EXPECT_MAT_NEAR(spmap_template, spmap, 0.0);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MeanShift, ALL_DEVICES);
////////////////////////////////////////////////////////////////////////////////
// MeanShiftSegmentation
namespace
{
IMPLEMENT_PARAM_CLASS(MinSize, int);
}
PARAM_TEST_CASE(MeanShiftSegmentation, cv::gpu::DeviceInfo, MinSize)
{
cv::gpu::DeviceInfo devInfo;
int minsize;
virtual void SetUp()
{
devInfo = GET_PARAM(0);
minsize = GET_PARAM(1);
cv::gpu::setDevice(devInfo.deviceID());
}
};
GPU_TEST_P(MeanShiftSegmentation, Regression)
{
cv::Mat img = readImageType("meanshift/cones.png", CV_8UC4);
ASSERT_FALSE(img.empty());
std::ostringstream path;
path << "meanshift/cones_segmented_sp10_sr10_minsize" << minsize;
if (supportFeature(devInfo, cv::gpu::FEATURE_SET_COMPUTE_20))
path << ".png";
else
path << "_CC1X.png";
cv::Mat dst_gold = readImage(path.str());
ASSERT_FALSE(dst_gold.empty());
cv::Mat dst;
cv::gpu::meanShiftSegmentation(loadMat(img), dst, 10, 10, minsize);
cv::Mat dst_rgb;
cv::cvtColor(dst, dst_rgb, cv::COLOR_BGRA2BGR);
EXPECT_MAT_SIMILAR(dst_gold, dst_rgb, 1e-3);
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MeanShiftSegmentation, testing::Combine(
ALL_DEVICES,
testing::Values(MinSize(0), MinSize(4), MinSize(20), MinSize(84), MinSize(340), MinSize(1364))));
#endif // HAVE_CUDA

43
modules/gpuimgproc/test/test_precomp.cpp
View File

@@ -1,43 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"

61
modules/gpuimgproc/test/test_precomp.hpp
View File

@@ -1,61 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifdef __GNUC__
# pragma GCC diagnostic ignored "-Wmissing-declarations"
# if defined __clang__ || defined __APPLE__
# pragma GCC diagnostic ignored "-Wmissing-prototypes"
# pragma GCC diagnostic ignored "-Wextra"
# endif
#endif
#ifndef __OPENCV_TEST_PRECOMP_HPP__
#define __OPENCV_TEST_PRECOMP_HPP__
#include "opencv2/ts.hpp"
#include "opencv2/ts/gpu_test.hpp"
#include "opencv2/gpuimgproc.hpp"
#include "opencv2/gpuarithm.hpp"
#include "opencv2/imgproc.hpp"
#endif