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

Merge branch 'master' into cuda_concurrency

Browse Source
This commit is contained in:
Ernest Galbrun 2014-07-31 16:03:51 +02:00
commit ae26368e78
95 changed files with 5076 additions and 10447 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
apps/traincascade/traincascade.cpp
View File

@ -13,6 +13,7 @@ int main( int argc, char* argv[] )
int numPos = 2000;
int numNeg = 1000;
int numStages = 20;
int numThreads = getNumThreads();
int precalcValBufSize = 256,
precalcIdxBufSize = 256;
bool baseFormatSave = false;
@ -36,6 +37,7 @@ int main( int argc, char* argv[] )
cout << " [-precalcValBufSize <precalculated_vals_buffer_size_in_Mb = " << precalcValBufSize << ">]" << endl;
cout << " [-precalcIdxBufSize <precalculated_idxs_buffer_size_in_Mb = " << precalcIdxBufSize << ">]" << endl;
cout << " [-baseFormatSave]" << endl;
cout << " [-numThreads <max_number_of_threads = " << numThreads << ">]" << endl;
cascadeParams.printDefaults();
stageParams.printDefaults();
for( int fi = 0; fi < fc; fi++ )
@ -82,6 +84,10 @@ int main( int argc, char* argv[] )
{
baseFormatSave = true;
}
else if( !strcmp( argv[i], "-numThreads" ) )
{
numThreads = atoi(argv[++i]);
}
else if ( cascadeParams.scanAttr( argv[i], argv[i+1] ) ) { i++; }
else if ( stageParams.scanAttr( argv[i], argv[i+1] ) ) { i++; }
else if ( !set )
@ -98,6 +104,7 @@ int main( int argc, char* argv[] )
}
}
setNumThreads( numThreads );
classifier.train( cascadeDirName,
vecName,
bgName,

6
cmake/OpenCVFindOpenNI2.cmake
View File

@ -21,13 +21,13 @@ if(WIN32)
find_library(OPENNI2_LIBRARY "OpenNI2" PATHS $ENV{OPENNI2_LIB64} DOC "OpenNI2 library")
endif()
elseif(UNIX OR APPLE)
find_file(OPENNI_INCLUDES "OpenNI.h" PATHS "/usr/include/ni2" "/usr/include/openni2" DOC "OpenNI2 c++ interface header")
find_library(OPENNI_LIBRARY "OpenNI2" PATHS "/usr/lib" DOC "OpenNI2 library")
find_file(OPENNI2_INCLUDES "OpenNI.h" PATHS "/usr/include/ni2" "/usr/include/openni2" DOC "OpenNI2 c++ interface header")
find_library(OPENNI2_LIBRARY "OpenNI2" PATHS "/usr/lib" DOC "OpenNI2 library")
endif()
if(OPENNI2_LIBRARY AND OPENNI2_INCLUDES)
set(HAVE_OPENNI2 TRUE)
endif() #if(OPENNI_LIBRARY AND OPENNI_INCLUDES)
endif() #if(OPENNI2_LIBRARY AND OPENNI2_INCLUDES)
get_filename_component(OPENNI2_LIB_DIR "${OPENNI2_LIBRARY}" PATH)
get_filename_component(OPENNI2_INCLUDE_DIR ${OPENNI2_INCLUDES} PATH)

161
doc/tutorials/features2d/akaze_matching/akaze_matching.rst Normal file
View File

@ -0,0 +1,161 @@
.. _akazeMatching:
AKAZE local features matching
******************************
Introduction
------------------
In this tutorial we will learn how to use [AKAZE]_ local features to detect and match keypoints on two images.
We will find keypoints on a pair of images with given homography matrix,
match them and count the number of inliers (i. e. matches that fit in the given homography).
You can find expanded version of this example here: https://github.com/pablofdezalc/test_kaze_akaze_opencv
.. [AKAZE] Fast Explicit Diffusion for Accelerated Features in Nonlinear Scale Spaces. Pablo F. Alcantarilla, Jesús Nuevo and Adrien Bartoli. In British Machine Vision Conference (BMVC), Bristol, UK, September 2013.
Data
------------------
We are going to use images 1 and 3 from *Graffity* sequence of Oxford dataset.
.. image:: images/graf.png
:height: 200pt
:width: 320pt
:alt: Graffity
:align: center
Homography is given by a 3 by 3 matrix:
.. code-block:: none
7.6285898e-01 -2.9922929e-01 2.2567123e+02
3.3443473e-01 1.0143901e+00 -7.6999973e+01
3.4663091e-04 -1.4364524e-05 1.0000000e+00
You can find the images (*graf1.png*, *graf3.png*) and homography (*H1to3p.xml*) in *opencv/samples/cpp*.
Source Code
===========
.. literalinclude:: ../../../../samples/cpp/tutorial_code/features2D/AKAZE_match.cpp
:language: cpp
:linenos:
:tab-width: 4
Explanation
===========
1. **Load images and homography**
.. code-block:: cpp
Mat img1 = imread("graf1.png", IMREAD_GRAYSCALE);
Mat img2 = imread("graf3.png", IMREAD_GRAYSCALE);
Mat homography;
FileStorage fs("H1to3p.xml", FileStorage::READ);
fs.getFirstTopLevelNode() >> homography;
We are loading grayscale images here. Homography is stored in the xml created with FileStorage.
2. **Detect keypoints and compute descriptors using AKAZE**
.. code-block:: cpp
vector<KeyPoint> kpts1, kpts2;
Mat desc1, desc2;
AKAZE akaze;
akaze(img1, noArray(), kpts1, desc1);
akaze(img2, noArray(), kpts2, desc2);
We create AKAZE object and use it's *operator()* functionality. Since we don't need the *mask* parameter, *noArray()* is used.
3. **Use brute-force matcher to find 2-nn matches**
.. code-block:: cpp
BFMatcher matcher(NORM_HAMMING);
vector< vector<DMatch> > nn_matches;
matcher.knnMatch(desc1, desc2, nn_matches, 2);
We use Hamming distance, because AKAZE uses binary descriptor by default.
4. **Use 2-nn matches to find correct keypoint matches**
.. code-block:: cpp
for(size_t i = 0; i < nn_matches.size(); i++) {
DMatch first = nn_matches[i][0];
float dist1 = nn_matches[i][0].distance;
float dist2 = nn_matches[i][1].distance;
if(dist1 < nn_match_ratio * dist2) {
matched1.push_back(kpts1[first.queryIdx]);
matched2.push_back(kpts2[first.trainIdx]);
}
}
If the closest match is *ratio* closer than the second closest one, then the match is correct.
5. **Check if our matches fit in the homography model**
.. code-block:: cpp
for(int i = 0; i < matched1.size(); i++) {
Mat col = Mat::ones(3, 1, CV_64F);
col.at<double>(0) = matched1[i].pt.x;
col.at<double>(1) = matched1[i].pt.y;
col = homography * col;
col /= col.at<double>(2);
float dist = sqrt( pow(col.at<double>(0) - matched2[i].pt.x, 2) +
pow(col.at<double>(1) - matched2[i].pt.y, 2));
if(dist < inlier_threshold) {
int new_i = inliers1.size();
inliers1.push_back(matched1[i]);
inliers2.push_back(matched2[i]);
good_matches.push_back(DMatch(new_i, new_i, 0));
}
}
If the distance from first keypoint's projection to the second keypoint is less than threshold, then it it fits in the homography.
We create a new set of matches for the inliers, because it is required by the drawing function.
6. **Output results**
.. code-block:: cpp
Mat res;
drawMatches(img1, inliers1, img2, inliers2, good_matches, res);
imwrite("res.png", res);
...
Here we save the resulting image and print some statistics.
Results
=======
Found matches
--------------
.. image:: images/res.png
:height: 200pt
:width: 320pt
:alt: Matches
:align: center
A-KAZE Matching Results
--------------------------
Keypoints 1: 2943
Keypoints 2: 3511
Matches: 447
Inliers: 308
Inliers Ratio: 0.689038

BIN
doc/tutorials/features2d/akaze_matching/images/graf.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 2.0 MiB

BIN
doc/tutorials/features2d/akaze_matching/images/res.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 1.8 MiB

BIN
doc/tutorials/features2d/table_of_content_features2d/images/AKAZE_Match_Tutorial_Cover.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 63 KiB

20
doc/tutorials/features2d/table_of_content_features2d/table_of_content_features2d.rst
View File

@ -183,6 +183,25 @@ Learn about how to use the feature points detectors, descriptors and matching f
:height: 90pt
:width: 90pt
+
.. tabularcolumns:: m{100pt} m{300pt}
.. cssclass:: toctableopencv
===================== ==============================================
|AkazeMatch| **Title:** :ref:`akazeMatching`
*Compatibility:* > OpenCV 3.0
*Author:* Fedor Morozov
Use *AKAZE* local features to find correspondence between two images.
===================== ==============================================
.. |AkazeMatch| image:: images/AKAZE_Match_Tutorial_Cover.png
:height: 90pt
:width: 90pt
.. raw:: latex
\pagebreak
@ -201,3 +220,4 @@ Learn about how to use the feature points detectors, descriptors and matching f
../feature_flann_matcher/feature_flann_matcher
../feature_homography/feature_homography
../detection_of_planar_objects/detection_of_planar_objects
../akaze_matching/akaze_matching

6
doc/user_guide/ug_traincascade.rst
View File

@ -200,6 +200,12 @@ Command line arguments of ``opencv_traincascade`` application grouped by purpose
This argument is actual in case of Haar-like features. If it is specified, the cascade will be saved in the old format.
* ``-numThreads <max_number_of_threads>``
Maximum number of threads to use during training. Notice that
the actual number of used threads may be lower, depending on
your machine and compilation options.
#.
Cascade parameters:

1
modules/core/include/opencv2/core/cvdef.h
View File

@ -244,6 +244,7 @@ typedef signed char schar;
/* fundamental constants */
#define CV_PI 3.1415926535897932384626433832795
#define CV_2PI 6.283185307179586476925286766559
#define CV_LOG2 0.69314718055994530941723212145818
/****************************************************************************************\

37
modules/core/perf/opencl/perf_dxt.cpp
View File

@ -54,23 +54,42 @@ namespace ocl {
///////////// dft ////////////////////////
typedef tuple<Size, int> DftParams;
enum OCL_FFT_TYPE
{
R2R = 0,
C2R = 1,
R2C = 2,
C2C = 3
};
typedef tuple<OCL_FFT_TYPE, Size, int> DftParams;
typedef TestBaseWithParam<DftParams> DftFixture;
OCL_PERF_TEST_P(DftFixture, Dft, ::testing::Combine(Values(OCL_SIZE_1, OCL_SIZE_2, OCL_SIZE_3),
Values((int)DFT_ROWS, (int)DFT_SCALE, (int)DFT_INVERSE,
(int)DFT_INVERSE | DFT_SCALE, (int)DFT_ROWS | DFT_INVERSE)))
OCL_PERF_TEST_P(DftFixture, Dft, ::testing::Combine(Values(C2C, R2R, C2R, R2C),
Values(OCL_SIZE_1, OCL_SIZE_2, OCL_SIZE_3, Size(512, 512), Size(1024, 1024), Size(2048, 2048)),
Values((int) 0, (int)DFT_ROWS, (int)DFT_SCALE, (int)DFT_INVERSE,
(int)DFT_INVERSE | DFT_SCALE, (int)DFT_ROWS | DFT_INVERSE)))
{
const DftParams params = GetParam();
const Size srcSize = get<0>(params);
const int flags = get<1>(params);
const int dft_type = get<0>(params);
const Size srcSize = get<1>(params);
int flags = get<2>(params);
UMat src(srcSize, CV_32FC2), dst(srcSize, CV_32FC2);
int in_cn, out_cn;
switch (dft_type)
{
case R2R: flags |= cv::DFT_REAL_OUTPUT; in_cn = 1; out_cn = 1; break;
case C2R: flags |= cv::DFT_REAL_OUTPUT; in_cn = 2; out_cn = 2; break;
case R2C: flags |= cv::DFT_COMPLEX_OUTPUT; in_cn = 1; out_cn = 2; break;
case C2C: flags |= cv::DFT_COMPLEX_OUTPUT; in_cn = 2; out_cn = 2; break;
}
UMat src(srcSize, CV_MAKE_TYPE(CV_32F, in_cn)), dst(srcSize, CV_MAKE_TYPE(CV_32F, out_cn));
declare.in(src, WARMUP_RNG).out(dst);
OCL_TEST_CYCLE() cv::dft(src, dst, flags | DFT_COMPLEX_OUTPUT);
OCL_TEST_CYCLE() cv::dft(src, dst, flags);
SANITY_CHECK(dst, 1e-3);
SANITY_CHECK(dst, 1e-5, ERROR_RELATIVE);
}
///////////// MulSpectrums ////////////////////////

109
modules/core/src/arithm.cpp
View File

@ -54,21 +54,23 @@ namespace cv
struct NOP {};
#if CV_SSE2
#if CV_SSE2 || CV_NEON
#define FUNCTOR_TEMPLATE(name) \
template<typename T> struct name {}
FUNCTOR_TEMPLATE(VLoadStore128);
#if CV_SSE2
FUNCTOR_TEMPLATE(VLoadStore64);
FUNCTOR_TEMPLATE(VLoadStore128Aligned);
#endif
#endif
template<typename T, class Op, class VOp>
void vBinOp(const T* src1, size_t step1, const T* src2, size_t step2, T* dst, size_t step, Size sz)
{
#if CV_SSE2
#if CV_SSE2 || CV_NEON
VOp vop;
#endif
Op op;
@ -79,9 +81,11 @@ void vBinOp(const T* src1, size_t step1, const T* src2, size_t step2, T* dst, si
{
int x = 0;
#if CV_NEON || CV_SSE2
#if CV_SSE2
if( USE_SSE2 )
{
#endif
for( ; x <= sz.width - 32/(int)sizeof(T); x += 32/sizeof(T) )
{
typename VLoadStore128<T>::reg_type r0 = VLoadStore128<T>::load(src1 + x );
@ -91,8 +95,10 @@ void vBinOp(const T* src1, size_t step1, const T* src2, size_t step2, T* dst, si
VLoadStore128<T>::store(dst + x , r0);
VLoadStore128<T>::store(dst + x + 16/sizeof(T), r1);
}
#if CV_SSE2
}
#endif
#endif
#if CV_SSE2
if( USE_SSE2 )
{
@ -125,7 +131,7 @@ template<typename T, class Op, class Op32>
void vBinOp32(const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, Size sz)
{
#if CV_SSE2
#if CV_SSE2 || CV_NEON
Op32 op32;
#endif
Op op;
@ -153,9 +159,11 @@ void vBinOp32(const T* src1, size_t step1, const T* src2, size_t step2,
}
}
#endif
#if CV_NEON || CV_SSE2
#if CV_SSE2
if( USE_SSE2 )
{
#endif
for( ; x <= sz.width - 8; x += 8 )
{
typename VLoadStore128<T>::reg_type r0 = VLoadStore128<T>::load(src1 + x );
@ -165,8 +173,10 @@ void vBinOp32(const T* src1, size_t step1, const T* src2, size_t step2,
VLoadStore128<T>::store(dst + x , r0);
VLoadStore128<T>::store(dst + x + 4, r1);
}
#if CV_SSE2
}
#endif
#endif
#if CV_ENABLE_UNROLLED
for( ; x <= sz.width - 4; x += 4 )
{
@ -383,7 +393,98 @@ FUNCTOR_TEMPLATE(VNot);
FUNCTOR_CLOSURE_1arg(VNot, uchar, return _mm_xor_si128(_mm_set1_epi32(-1), a));
#endif
#if CV_SSE2
#if CV_NEON
#define FUNCTOR_LOADSTORE(name, template_arg, register_type, load_body, store_body)\
template <> \
struct name<template_arg>{ \
typedef register_type reg_type; \
static reg_type load(const template_arg * p) { return load_body (p);}; \
static void store(template_arg * p, reg_type v) { store_body (p, v);}; \
}
#define FUNCTOR_CLOSURE_2arg(name, template_arg, body)\
template<> \
struct name<template_arg> \
{ \
VLoadStore128<template_arg>::reg_type operator()( \
VLoadStore128<template_arg>::reg_type a, \
VLoadStore128<template_arg>::reg_type b) const \
{ \
return body; \
}; \
}
#define FUNCTOR_CLOSURE_1arg(name, template_arg, body)\
template<> \
struct name<template_arg> \
{ \
VLoadStore128<template_arg>::reg_type operator()( \
VLoadStore128<template_arg>::reg_type a, \
VLoadStore128<template_arg>::reg_type ) const \
{ \
return body; \
}; \
}
FUNCTOR_LOADSTORE(VLoadStore128, uchar, uint8x16_t, vld1q_u8 , vst1q_u8 );
FUNCTOR_LOADSTORE(VLoadStore128, schar, int8x16_t, vld1q_s8 , vst1q_s8 );
FUNCTOR_LOADSTORE(VLoadStore128, ushort, uint16x8_t, vld1q_u16, vst1q_u16);
FUNCTOR_LOADSTORE(VLoadStore128, short, int16x8_t, vld1q_s16, vst1q_s16);
FUNCTOR_LOADSTORE(VLoadStore128, int, int32x4_t, vld1q_s32, vst1q_s32);
FUNCTOR_LOADSTORE(VLoadStore128, float, float32x4_t, vld1q_f32, vst1q_f32);
FUNCTOR_TEMPLATE(VAdd);
FUNCTOR_CLOSURE_2arg(VAdd, uchar, vqaddq_u8 (a, b));
FUNCTOR_CLOSURE_2arg(VAdd, schar, vqaddq_s8 (a, b));
FUNCTOR_CLOSURE_2arg(VAdd, ushort, vqaddq_u16(a, b));
FUNCTOR_CLOSURE_2arg(VAdd, short, vqaddq_s16(a, b));
FUNCTOR_CLOSURE_2arg(VAdd, int, vaddq_s32 (a, b));
FUNCTOR_CLOSURE_2arg(VAdd, float, vaddq_f32 (a, b));
FUNCTOR_TEMPLATE(VSub);
FUNCTOR_CLOSURE_2arg(VSub, uchar, vqsubq_u8 (a, b));
FUNCTOR_CLOSURE_2arg(VSub, schar, vqsubq_s8 (a, b));
FUNCTOR_CLOSURE_2arg(VSub, ushort, vqsubq_u16(a, b));
FUNCTOR_CLOSURE_2arg(VSub, short, vqsubq_s16(a, b));
FUNCTOR_CLOSURE_2arg(VSub, int, vsubq_s32 (a, b));
FUNCTOR_CLOSURE_2arg(VSub, float, vsubq_f32 (a, b));
FUNCTOR_TEMPLATE(VMin);
FUNCTOR_CLOSURE_2arg(VMin, uchar, vminq_u8 (a, b));
FUNCTOR_CLOSURE_2arg(VMin, schar, vminq_s8 (a, b));
FUNCTOR_CLOSURE_2arg(VMin, ushort, vminq_u16(a, b));
FUNCTOR_CLOSURE_2arg(VMin, short, vminq_s16(a, b));
FUNCTOR_CLOSURE_2arg(VMin, int, vminq_s32(a, b));
FUNCTOR_CLOSURE_2arg(VMin, float, vminq_f32(a, b));
FUNCTOR_TEMPLATE(VMax);
FUNCTOR_CLOSURE_2arg(VMax, uchar, vmaxq_u8 (a, b));
FUNCTOR_CLOSURE_2arg(VMax, schar, vmaxq_s8 (a, b));
FUNCTOR_CLOSURE_2arg(VMax, ushort, vmaxq_u16(a, b));
FUNCTOR_CLOSURE_2arg(VMax, short, vmaxq_s16(a, b));
FUNCTOR_CLOSURE_2arg(VMax, int, vmaxq_s32(a, b));
FUNCTOR_CLOSURE_2arg(VMax, float, vmaxq_f32(a, b));
FUNCTOR_TEMPLATE(VAbsDiff);
FUNCTOR_CLOSURE_2arg(VAbsDiff, uchar, vabdq_u8 (a, b));
FUNCTOR_CLOSURE_2arg(VAbsDiff, schar, vqabsq_s8 (vqsubq_s8(a, b)));
FUNCTOR_CLOSURE_2arg(VAbsDiff, ushort, vabdq_u16 (a, b));
FUNCTOR_CLOSURE_2arg(VAbsDiff, short, vqabsq_s16(vqsubq_s16(a, b)));
FUNCTOR_CLOSURE_2arg(VAbsDiff, int, vabdq_s32 (a, b));
FUNCTOR_CLOSURE_2arg(VAbsDiff, float, vabdq_f32 (a, b));
FUNCTOR_TEMPLATE(VAnd);
FUNCTOR_CLOSURE_2arg(VAnd, uchar, vandq_u8(a, b));
FUNCTOR_TEMPLATE(VOr);
FUNCTOR_CLOSURE_2arg(VOr , uchar, vorrq_u8(a, b));
FUNCTOR_TEMPLATE(VXor);
FUNCTOR_CLOSURE_2arg(VXor, uchar, veorq_u8(a, b));
FUNCTOR_TEMPLATE(VNot);
FUNCTOR_CLOSURE_1arg(VNot, uchar, vmvnq_u8(a ));
#endif
#if CV_SSE2 || CV_NEON
#define IF_SIMD(op) op
#else
#define IF_SIMD(op) NOP

396
modules/core/src/dxt.cpp
View File

@ -43,6 +43,7 @@
#include "opencv2/core/opencl/runtime/opencl_clamdfft.hpp"
#include "opencv2/core/opencl/runtime/opencl_core.hpp"
#include "opencl_kernels.hpp"
#include <map>
namespace cv
{
@ -1781,6 +1782,375 @@ static bool ippi_DFT_R_32F(const Mat& src, Mat& dst, bool inv, int norm_flag)
#endif
}
#ifdef HAVE_OPENCL
namespace cv
{
enum FftType
{
R2R = 0, // real to CCS in case forward transform, CCS to real otherwise
C2R = 1, // complex to real in case inverse transform
R2C = 2, // real to complex in case forward transform
C2C = 3 // complex to complex
};
struct OCL_FftPlan
{
private:
UMat twiddles;
String buildOptions;
int thread_count;
bool status;
int dft_size;
public:
OCL_FftPlan(int _size): dft_size(_size), status(true)
{
int min_radix;
std::vector<int> radixes, blocks;
ocl_getRadixes(dft_size, radixes, blocks, min_radix);
thread_count = dft_size / min_radix;
if (thread_count > (int) ocl::Device::getDefault().maxWorkGroupSize())
{
status = false;
return;
}
// generate string with radix calls
String radix_processing;
int n = 1, twiddle_size = 0;
for (size_t i=0; i<radixes.size(); i++)
{
int radix = radixes[i], block = blocks[i];
if (block > 1)
radix_processing += format("fft_radix%d_B%d(smem,twiddles+%d,ind,%d,%d);", radix, block, twiddle_size, n, dft_size/radix);
else
radix_processing += format("fft_radix%d(smem,twiddles+%d,ind,%d,%d);", radix, twiddle_size, n, dft_size/radix);
twiddle_size += (radix-1)*n;
n *= radix;
}
Mat tw(1, twiddle_size, CV_32FC2);
float* ptr = tw.ptr<float>();
int ptr_index = 0;
n = 1;
for (size_t i=0; i<radixes.size(); i++)
{
int radix = radixes[i];
n *= radix;
for (int j=1; j<radix; j++)
{
double theta = -CV_2PI*j/n;
for (int k=0; k<(n/radix); k++)
{
ptr[ptr_index++] = (float) cos(k*theta);
ptr[ptr_index++] = (float) sin(k*theta);
}
}
}
twiddles = tw.getUMat(ACCESS_READ);
buildOptions = format("-D LOCAL_SIZE=%d -D kercn=%d -D RADIX_PROCESS=%s",
dft_size, min_radix, radix_processing.c_str());
}
bool enqueueTransform(InputArray _src, OutputArray _dst, int num_dfts, int flags, int fftType, bool rows = true) const
{
if (!status)
return false;
UMat src = _src.getUMat();
UMat dst = _dst.getUMat();
size_t globalsize[2];
size_t localsize[2];
String kernel_name;
bool is1d = (flags & DFT_ROWS) != 0 || num_dfts == 1;
bool inv = (flags & DFT_INVERSE) != 0;
String options = buildOptions;
if (rows)
{
globalsize[0] = thread_count; globalsize[1] = src.rows;
localsize[0] = thread_count; localsize[1] = 1;
kernel_name = !inv ? "fft_multi_radix_rows" : "ifft_multi_radix_rows";
if ((is1d || inv) && (flags & DFT_SCALE))
options += " -D DFT_SCALE";
}
else
{
globalsize[0] = num_dfts; globalsize[1] = thread_count;
localsize[0] = 1; localsize[1] = thread_count;
kernel_name = !inv ? "fft_multi_radix_cols" : "ifft_multi_radix_cols";
if (flags & DFT_SCALE)
options += " -D DFT_SCALE";
}
options += src.channels() == 1 ? " -D REAL_INPUT" : " -D COMPLEX_INPUT";
options += dst.channels() == 1 ? " -D REAL_OUTPUT" : " -D COMPLEX_OUTPUT";
options += is1d ? " -D IS_1D" : "";
if (!inv)
{
if ((is1d && src.channels() == 1) || (rows && (fftType == R2R)))
options += " -D NO_CONJUGATE";
}
else
{
if (rows && (fftType == C2R || fftType == R2R))
options += " -D NO_CONJUGATE";
if (dst.cols % 2 == 0)
options += " -D EVEN";
}
ocl::Kernel k(kernel_name.c_str(), ocl::core::fft_oclsrc, options);
if (k.empty())
return false;
k.args(ocl::KernelArg::ReadOnly(src), ocl::KernelArg::WriteOnly(dst), ocl::KernelArg::PtrReadOnly(twiddles), thread_count, num_dfts);
return k.run(2, globalsize, localsize, false);
}
private:
static void ocl_getRadixes(int cols, std::vector<int>& radixes, std::vector<int>& blocks, int& min_radix)
{
int factors[34];
int nf = DFTFactorize(cols, factors);
int n = 1;
int factor_index = 0;
min_radix = INT_MAX;
// 2^n transforms
if ((factors[factor_index] & 1) == 0)
{
for( ; n < factors[factor_index];)
{
int radix = 2, block = 1;
if (8*n <= factors[0])
radix = 8;
else if (4*n <= factors[0])
{
radix = 4;
if (cols % 12 == 0)
block = 3;
else if (cols % 8 == 0)
block = 2;
}
else
{
if (cols % 10 == 0)
block = 5;
else if (cols % 8 == 0)
block = 4;
else if (cols % 6 == 0)
block = 3;
else if (cols % 4 == 0)
block = 2;
}
radixes.push_back(radix);
blocks.push_back(block);
min_radix = min(min_radix, block*radix);
n *= radix;
}
factor_index++;
}
// all the other transforms
for( ; factor_index < nf; factor_index++)
{
int radix = factors[factor_index], block = 1;
if (radix == 3)
{
if (cols % 12 == 0)
block = 4;
else if (cols % 9 == 0)
block = 3;
else if (cols % 6 == 0)
block = 2;
}
else if (radix == 5)
{
if (cols % 10 == 0)
block = 2;
}
radixes.push_back(radix);
blocks.push_back(block);
min_radix = min(min_radix, block*radix);
}
}
};
class OCL_FftPlanCache
{
public:
static OCL_FftPlanCache & getInstance()
{
static OCL_FftPlanCache planCache;
return planCache;
}
Ptr<OCL_FftPlan> getFftPlan(int dft_size)
{
std::map<int, Ptr<OCL_FftPlan> >::iterator f = planStorage.find(dft_size);
if (f != planStorage.end())
{
return f->second;
}
else
{
Ptr<OCL_FftPlan> newPlan = Ptr<OCL_FftPlan>(new OCL_FftPlan(dft_size));
planStorage[dft_size] = newPlan;
return newPlan;
}
}
~OCL_FftPlanCache()
{
planStorage.clear();
}
protected:
OCL_FftPlanCache() :
planStorage()
{
}
std::map<int, Ptr<OCL_FftPlan> > planStorage;
};
static bool ocl_dft_rows(InputArray _src, OutputArray _dst, int nonzero_rows, int flags, int fftType)
{
Ptr<OCL_FftPlan> plan = OCL_FftPlanCache::getInstance().getFftPlan(_src.cols());
return plan->enqueueTransform(_src, _dst, nonzero_rows, flags, fftType, true);
}
static bool ocl_dft_cols(InputArray _src, OutputArray _dst, int nonzero_cols, int flags, int fftType)
{
Ptr<OCL_FftPlan> plan = OCL_FftPlanCache::getInstance().getFftPlan(_src.rows());
return plan->enqueueTransform(_src, _dst, nonzero_cols, flags, fftType, false);
}
static bool ocl_dft(InputArray _src, OutputArray _dst, int flags, int nonzero_rows)
{
int type = _src.type(), cn = CV_MAT_CN(type);
Size ssize = _src.size();
if ( !(type == CV_32FC1 || type == CV_32FC2) )
return false;
// if is not a multiplication of prime numbers { 2, 3, 5 }
if (ssize.area() != getOptimalDFTSize(ssize.area()))
return false;
UMat src = _src.getUMat();
int complex_input = cn == 2 ? 1 : 0;
int complex_output = (flags & DFT_COMPLEX_OUTPUT) != 0;
int real_input = cn == 1 ? 1 : 0;
int real_output = (flags & DFT_REAL_OUTPUT) != 0;
bool inv = (flags & DFT_INVERSE) != 0 ? 1 : 0;
if( nonzero_rows <= 0 || nonzero_rows > _src.rows() )
nonzero_rows = _src.rows();
bool is1d = (flags & DFT_ROWS) != 0 || nonzero_rows == 1;
// if output format is not specified
if (complex_output + real_output == 0)
{
if (real_input)
real_output = 1;
else
complex_output = 1;
}
FftType fftType = (FftType)(complex_input << 0 | complex_output << 1);
// Forward Complex to CCS not supported
if (fftType == C2R && !inv)
fftType = C2C;
// Inverse CCS to Complex not supported
if (fftType == R2C && inv)
fftType = R2R;
UMat output;
if (fftType == C2C || fftType == R2C)
{
// complex output
_dst.create(src.size(), CV_32FC2);
output = _dst.getUMat();
}
else
{
// real output
if (is1d)
{
_dst.create(src.size(), CV_32FC1);
output = _dst.getUMat();
}
else
{
_dst.create(src.size(), CV_32FC1);
output.create(src.size(), CV_32FC2);
}
}
if (!inv)
{
if (!ocl_dft_rows(src, output, nonzero_rows, flags, fftType))
return false;
if (!is1d)
{
int nonzero_cols = fftType == R2R ? output.cols/2 + 1 : output.cols;
if (!ocl_dft_cols(output, _dst, nonzero_cols, flags, fftType))
return false;
}
}
else
{
if (fftType == C2C)
{
// complex output
if (!ocl_dft_rows(src, output, nonzero_rows, flags, fftType))
return false;
if (!is1d)
{
if (!ocl_dft_cols(output, output, output.cols, flags, fftType))
return false;
}
}
else
{
if (is1d)
{
if (!ocl_dft_rows(src, output, nonzero_rows, flags, fftType))
return false;
}
else
{
int nonzero_cols = src.cols/2 + 1;
if (!ocl_dft_cols(src, output, nonzero_cols, flags, fftType))
return false;
if (!ocl_dft_rows(output, _dst, nonzero_rows, flags, fftType))
return false;
}
}
}
return true;
}
} // namespace cv;
#endif
#ifdef HAVE_CLAMDFFT
namespace cv {
@ -1791,14 +2161,6 @@ namespace cv {
CV_Assert(s == CLFFT_SUCCESS); \
}
enum FftType
{
R2R = 0, // real to real
C2R = 1, // opencl HERMITIAN_INTERLEAVED to real
R2C = 2, // real to opencl HERMITIAN_INTERLEAVED
C2C = 3 // complex to complex
};
class PlanCache
{
struct FftPlan
@ -1923,7 +2285,7 @@ public:
}
// no baked plan is found, so let's create a new one
FftPlan * newPlan = new FftPlan(dft_size, src_step, dst_step, doubleFP, inplace, flags, fftType);
Ptr<FftPlan> newPlan = Ptr<FftPlan>(new FftPlan(dft_size, src_step, dst_step, doubleFP, inplace, flags, fftType));
planStorage.push_back(newPlan);
return newPlan->plHandle;
@ -1931,8 +2293,6 @@ public:
~PlanCache()
{
for (std::vector<FftPlan *>::iterator i = planStorage.begin(), end = planStorage.end(); i != end; ++i)
delete (*i);
planStorage.clear();
}
@ -1942,7 +2302,7 @@ protected:
{
}
std::vector<FftPlan *> planStorage;
std::vector<Ptr<FftPlan> > planStorage;
};
extern "C" {
@ -1960,7 +2320,7 @@ static void CL_CALLBACK oclCleanupCallback(cl_event e, cl_int, void *p)
}
static bool ocl_dft(InputArray _src, OutputArray _dst, int flags)
static bool ocl_dft_amdfft(InputArray _src, OutputArray _dst, int flags)
{
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
Size ssize = _src.size();
@ -2019,7 +2379,6 @@ static bool ocl_dft(InputArray _src, OutputArray _dst, int flags)
tmpBuffer.addref();
clSetEventCallback(e, CL_COMPLETE, oclCleanupCallback, tmpBuffer.u);
return true;
}
@ -2034,7 +2393,12 @@ void cv::dft( InputArray _src0, OutputArray _dst, int flags, int nonzero_rows )
#ifdef HAVE_CLAMDFFT
CV_OCL_RUN(ocl::haveAmdFft() && ocl::Device::getDefault().type() != ocl::Device::TYPE_CPU &&
_dst.isUMat() && _src0.dims() <= 2 && nonzero_rows == 0,
ocl_dft(_src0, _dst, flags))
ocl_dft_amdfft(_src0, _dst, flags))
#endif
#ifdef HAVE_OPENCL
CV_OCL_RUN(_dst.isUMat() && _src0.dims() <= 2,
ocl_dft(_src0, _dst, flags, nonzero_rows))
#endif
static DFTFunc dft_tbl[6] =
@ -2046,10 +2410,8 @@ void cv::dft( InputArray _src0, OutputArray _dst, int flags, int nonzero_rows )
(DFTFunc)RealDFT_64f,
(DFTFunc)CCSIDFT_64f
};
AutoBuffer<uchar> buf;
void *spec = 0;
Mat src0 = _src0.getMat(), src = src0;
int prev_len = 0, stage = 0;
bool inv = (flags & DFT_INVERSE) != 0;

864
modules/core/src/opencl/fft.cl Normal file
View File

@ -0,0 +1,864 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
// Copyright (C) 2014, Itseez, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
#define SQRT_2 0.707106781188f
#define sin_120 0.866025403784f
#define fft5_2 0.559016994374f
#define fft5_3 -0.951056516295f
#define fft5_4 -1.538841768587f
#define fft5_5 0.363271264002f
__attribute__((always_inline))
float2 mul_float2(float2 a, float2 b) {
return (float2)(fma(a.x, b.x, -a.y * b.y), fma(a.x, b.y, a.y * b.x));
}
__attribute__((always_inline))
float2 twiddle(float2 a) {
return (float2)(a.y, -a.x);
}
__attribute__((always_inline))
void butterfly2(float2 a0, float2 a1, __local float2* smem, __global const float2* twiddles,
const int x, const int block_size)
{
const int k = x & (block_size - 1);
a1 = mul_float2(twiddles[k], a1);
const int dst_ind = (x << 1) - k;
smem[dst_ind] = a0 + a1;
smem[dst_ind+block_size] = a0 - a1;
}
__attribute__((always_inline))
void butterfly4(float2 a0, float2 a1, float2 a2, float2 a3, __local float2* smem, __global const float2* twiddles,
const int x, const int block_size)
{
const int k = x & (block_size - 1);
a1 = mul_float2(twiddles[k], a1);
a2 = mul_float2(twiddles[k + block_size], a2);
a3 = mul_float2(twiddles[k + 2*block_size], a3);
const int dst_ind = ((x - k) << 2) + k;
float2 b0 = a0 + a2;
a2 = a0 - a2;
float2 b1 = a1 + a3;
a3 = twiddle(a1 - a3);
smem[dst_ind] = b0 + b1;
smem[dst_ind + block_size] = a2 + a3;
smem[dst_ind + 2*block_size] = b0 - b1;
smem[dst_ind + 3*block_size] = a2 - a3;
}
__attribute__((always_inline))
void butterfly3(float2 a0, float2 a1, float2 a2, __local float2* smem, __global const float2* twiddles,
const int x, const int block_size)
{
const int k = x % block_size;
a1 = mul_float2(twiddles[k], a1);
a2 = mul_float2(twiddles[k+block_size], a2);
const int dst_ind = ((x - k) * 3) + k;
float2 b1 = a1 + a2;
a2 = twiddle(sin_120*(a1 - a2));
float2 b0 = a0 - (float2)(0.5f)*b1;
smem[dst_ind] = a0 + b1;
smem[dst_ind + block_size] = b0 + a2;
smem[dst_ind + 2*block_size] = b0 - a2;
}
__attribute__((always_inline))
void butterfly5(float2 a0, float2 a1, float2 a2, float2 a3, float2 a4, __local float2* smem, __global const float2* twiddles,
const int x, const int block_size)
{
const int k = x % block_size;
a1 = mul_float2(twiddles[k], a1);
a2 = mul_float2(twiddles[k + block_size], a2);
a3 = mul_float2(twiddles[k+2*block_size], a3);
a4 = mul_float2(twiddles[k+3*block_size], a4);
const int dst_ind = ((x - k) * 5) + k;
__local float2* dst = smem + dst_ind;
float2 b0, b1, b5;
b1 = a1 + a4;
a1 -= a4;
a4 = a3 + a2;
a3 -= a2;
a2 = b1 + a4;
b0 = a0 - (float2)0.25f * a2;
b1 = fft5_2 * (b1 - a4);
a4 = fft5_3 * (float2)(-a1.y - a3.y, a1.x + a3.x);
b5 = (float2)(a4.x - fft5_5 * a1.y, a4.y + fft5_5 * a1.x);
a4.x += fft5_4 * a3.y;
a4.y -= fft5_4 * a3.x;
a1 = b0 + b1;
b0 -= b1;
dst[0] = a0 + a2;
dst[block_size] = a1 + a4;
dst[2 * block_size] = b0 + b5;
dst[3 * block_size] = b0 - b5;
dst[4 * block_size] = a1 - a4;
}
__attribute__((always_inline))
void fft_radix2(__local float2* smem, __global const float2* twiddles, const int x, const int block_size, const int t)
{
float2 a0, a1;
if (x < t)
{
a0 = smem[x];
a1 = smem[x+t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x < t)
butterfly2(a0, a1, smem, twiddles, x, block_size);
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix2_B2(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1 + t/2;
float2 a0, a1, a2, a3;
if (x1 < t/2)
{
a0 = smem[x1]; a1 = smem[x1+t];
a2 = smem[x2]; a3 = smem[x2+t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/2)
{
butterfly2(a0, a1, smem, twiddles, x1, block_size);
butterfly2(a2, a3, smem, twiddles, x2, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix2_B3(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1 + t/3;
const int x3 = x1 + 2*t/3;
float2 a0, a1, a2, a3, a4, a5;
if (x1 < t/3)
{
a0 = smem[x1]; a1 = smem[x1+t];
a2 = smem[x2]; a3 = smem[x2+t];
a4 = smem[x3]; a5 = smem[x3+t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/3)
{
butterfly2(a0, a1, smem, twiddles, x1, block_size);
butterfly2(a2, a3, smem, twiddles, x2, block_size);
butterfly2(a4, a5, smem, twiddles, x3, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix2_B4(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int thread_block = t/4;
const int x2 = x1 + thread_block;
const int x3 = x1 + 2*thread_block;
const int x4 = x1 + 3*thread_block;
float2 a0, a1, a2, a3, a4, a5, a6, a7;
if (x1 < t/4)
{
a0 = smem[x1]; a1 = smem[x1+t];
a2 = smem[x2]; a3 = smem[x2+t];
a4 = smem[x3]; a5 = smem[x3+t];
a6 = smem[x4]; a7 = smem[x4+t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/4)
{
butterfly2(a0, a1, smem, twiddles, x1, block_size);
butterfly2(a2, a3, smem, twiddles, x2, block_size);
butterfly2(a4, a5, smem, twiddles, x3, block_size);
butterfly2(a6, a7, smem, twiddles, x4, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix2_B5(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int thread_block = t/5;
const int x2 = x1 + thread_block;
const int x3 = x1 + 2*thread_block;
const int x4 = x1 + 3*thread_block;
const int x5 = x1 + 4*thread_block;
float2 a0, a1, a2, a3, a4, a5, a6, a7, a8, a9;
if (x1 < t/5)
{
a0 = smem[x1]; a1 = smem[x1+t];
a2 = smem[x2]; a3 = smem[x2+t];
a4 = smem[x3]; a5 = smem[x3+t];
a6 = smem[x4]; a7 = smem[x4+t];
a8 = smem[x5]; a9 = smem[x5+t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/5)
{
butterfly2(a0, a1, smem, twiddles, x1, block_size);
butterfly2(a2, a3, smem, twiddles, x2, block_size);
butterfly2(a4, a5, smem, twiddles, x3, block_size);
butterfly2(a6, a7, smem, twiddles, x4, block_size);
butterfly2(a8, a9, smem, twiddles, x5, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix4(__local float2* smem, __global const float2* twiddles, const int x, const int block_size, const int t)
{
float2 a0, a1, a2, a3;
if (x < t)
{
a0 = smem[x]; a1 = smem[x+t]; a2 = smem[x+2*t]; a3 = smem[x+3*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x < t)
butterfly4(a0, a1, a2, a3, smem, twiddles, x, block_size);
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix4_B2(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1 + t/2;
float2 a0, a1, a2, a3, a4, a5, a6, a7;
if (x1 < t/2)
{
a0 = smem[x1]; a1 = smem[x1+t]; a2 = smem[x1+2*t]; a3 = smem[x1+3*t];
a4 = smem[x2]; a5 = smem[x2+t]; a6 = smem[x2+2*t]; a7 = smem[x2+3*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/2)
{
butterfly4(a0, a1, a2, a3, smem, twiddles, x1, block_size);
butterfly4(a4, a5, a6, a7, smem, twiddles, x2, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix4_B3(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1 + t/3;
const int x3 = x2 + t/3;
float2 a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11;
if (x1 < t/3)
{
a0 = smem[x1]; a1 = smem[x1+t]; a2 = smem[x1+2*t]; a3 = smem[x1+3*t];
a4 = smem[x2]; a5 = smem[x2+t]; a6 = smem[x2+2*t]; a7 = smem[x2+3*t];
a8 = smem[x3]; a9 = smem[x3+t]; a10 = smem[x3+2*t]; a11 = smem[x3+3*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/3)
{
butterfly4(a0, a1, a2, a3, smem, twiddles, x1, block_size);
butterfly4(a4, a5, a6, a7, smem, twiddles, x2, block_size);
butterfly4(a8, a9, a10, a11, smem, twiddles, x3, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix8(__local float2* smem, __global const float2* twiddles, const int x, const int block_size, const int t)
{
const int k = x % block_size;
float2 a0, a1, a2, a3, a4, a5, a6, a7;
if (x < t)
{
int tw_ind = block_size / 8;
a0 = smem[x];
a1 = mul_float2(twiddles[k], smem[x + t]);
a2 = mul_float2(twiddles[k + block_size],smem[x+2*t]);
a3 = mul_float2(twiddles[k+2*block_size],smem[x+3*t]);
a4 = mul_float2(twiddles[k+3*block_size],smem[x+4*t]);
a5 = mul_float2(twiddles[k+4*block_size],smem[x+5*t]);
a6 = mul_float2(twiddles[k+5*block_size],smem[x+6*t]);
a7 = mul_float2(twiddles[k+6*block_size],smem[x+7*t]);
float2 b0, b1, b6, b7;
b0 = a0 + a4;
a4 = a0 - a4;
b1 = a1 + a5;
a5 = a1 - a5;
a5 = (float2)(SQRT_2) * (float2)(a5.x + a5.y, -a5.x + a5.y);
b6 = twiddle(a2 - a6);
a2 = a2 + a6;
b7 = a3 - a7;
b7 = (float2)(SQRT_2) * (float2)(-b7.x + b7.y, -b7.x - b7.y);
a3 = a3 + a7;
a0 = b0 + a2;
a2 = b0 - a2;
a1 = b1 + a3;
a3 = twiddle(b1 - a3);
a6 = a4 - b6;
a4 = a4 + b6;
a7 = twiddle(a5 - b7);
a5 = a5 + b7;
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x < t)
{
const int dst_ind = ((x - k) << 3) + k;
__local float2* dst = smem + dst_ind;
dst[0] = a0 + a1;
dst[block_size] = a4 + a5;
dst[2 * block_size] = a2 + a3;
dst[3 * block_size] = a6 + a7;
dst[4 * block_size] = a0 - a1;
dst[5 * block_size] = a4 - a5;
dst[6 * block_size] = a2 - a3;
dst[7 * block_size] = a6 - a7;
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix3(__local float2* smem, __global const float2* twiddles, const int x, const int block_size, const int t)
{
float2 a0, a1, a2;
if (x < t)
{
a0 = smem[x]; a1 = smem[x+t]; a2 = smem[x+2*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x < t)
butterfly3(a0, a1, a2, smem, twiddles, x, block_size);
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix3_B2(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1 + t/2;
float2 a0, a1, a2, a3, a4, a5;
if (x1 < t/2)
{
a0 = smem[x1]; a1 = smem[x1+t]; a2 = smem[x1+2*t];
a3 = smem[x2]; a4 = smem[x2+t]; a5 = smem[x2+2*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/2)
{
butterfly3(a0, a1, a2, smem, twiddles, x1, block_size);
butterfly3(a3, a4, a5, smem, twiddles, x2, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix3_B3(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1 + t/3;
const int x3 = x2 + t/3;
float2 a0, a1, a2, a3, a4, a5, a6, a7, a8;
if (x1 < t/2)
{
a0 = smem[x1]; a1 = smem[x1+t]; a2 = smem[x1+2*t];
a3 = smem[x2]; a4 = smem[x2+t]; a5 = smem[x2+2*t];
a6 = smem[x3]; a7 = smem[x3+t]; a8 = smem[x3+2*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/2)
{
butterfly3(a0, a1, a2, smem, twiddles, x1, block_size);
butterfly3(a3, a4, a5, smem, twiddles, x2, block_size);
butterfly3(a6, a7, a8, smem, twiddles, x3, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix3_B4(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int thread_block = t/4;
const int x2 = x1 + thread_block;
const int x3 = x1 + 2*thread_block;
const int x4 = x1 + 3*thread_block;
float2 a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11;
if (x1 < t/4)
{
a0 = smem[x1]; a1 = smem[x1+t]; a2 = smem[x1+2*t];
a3 = smem[x2]; a4 = smem[x2+t]; a5 = smem[x2+2*t];
a6 = smem[x3]; a7 = smem[x3+t]; a8 = smem[x3+2*t];
a9 = smem[x4]; a10 = smem[x4+t]; a11 = smem[x4+2*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/4)
{
butterfly3(a0, a1, a2, smem, twiddles, x1, block_size);
butterfly3(a3, a4, a5, smem, twiddles, x2, block_size);
butterfly3(a6, a7, a8, smem, twiddles, x3, block_size);
butterfly3(a9, a10, a11, smem, twiddles, x4, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix5(__local float2* smem, __global const float2* twiddles, const int x, const int block_size, const int t)
{
const int k = x % block_size;
float2 a0, a1, a2, a3, a4;
if (x < t)
{
a0 = smem[x]; a1 = smem[x + t]; a2 = smem[x+2*t]; a3 = smem[x+3*t]; a4 = smem[x+4*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x < t)
butterfly5(a0, a1, a2, a3, a4, smem, twiddles, x, block_size);
barrier(CLK_LOCAL_MEM_FENCE);
}
__attribute__((always_inline))
void fft_radix5_B2(__local float2* smem, __global const float2* twiddles, const int x1, const int block_size, const int t)
{
const int x2 = x1+t/2;
float2 a0, a1, a2, a3, a4, a5, a6, a7, a8, a9;
if (x1 < t/2)
{
a0 = smem[x1]; a1 = smem[x1 + t]; a2 = smem[x1+2*t]; a3 = smem[x1+3*t]; a4 = smem[x1+4*t];
a5 = smem[x2]; a6 = smem[x2 + t]; a7 = smem[x2+2*t]; a8 = smem[x2+3*t]; a9 = smem[x2+4*t];
}
barrier(CLK_LOCAL_MEM_FENCE);
if (x1 < t/2)
{
butterfly5(a0, a1, a2, a3, a4, smem, twiddles, x1, block_size);
butterfly5(a5, a6, a7, a8, a9, smem, twiddles, x2, block_size);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
#ifdef DFT_SCALE
#define SCALE_VAL(x, scale) x*scale
#else
#define SCALE_VAL(x, scale) x
#endif
__kernel void fft_multi_radix_rows(__global const uchar* src_ptr, int src_step, int src_offset, int src_rows, int src_cols,
__global uchar* dst_ptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
__global float2* twiddles_ptr, const int t, const int nz)
{
const int x = get_global_id(0);
const int y = get_group_id(1);
const int block_size = LOCAL_SIZE/kercn;
if (y < nz)
{
__local float2 smem[LOCAL_SIZE];
__global const float2* twiddles = (__global float2*) twiddles_ptr;
const int ind = x;
#ifdef IS_1D
float scale = 1.f/dst_cols;
#else
float scale = 1.f/(dst_cols*dst_rows);
#endif
#ifdef COMPLEX_INPUT
__global const float2* src = (__global const float2*)(src_ptr + mad24(y, src_step, mad24(x, (int)(sizeof(float)*2), src_offset)));
#pragma unroll
for (int i=0; i<kercn; i++)
smem[x+i*block_size] = src[i*block_size];
#else
__global const float* src = (__global const float*)(src_ptr + mad24(y, src_step, mad24(x, (int)sizeof(float), src_offset)));
#pragma unroll
for (int i=0; i<kercn; i++)
smem[x+i*block_size] = (float2)(src[i*block_size], 0.f);
#endif
barrier(CLK_LOCAL_MEM_FENCE);
RADIX_PROCESS;
#ifdef COMPLEX_OUTPUT
#ifdef NO_CONJUGATE
// copy result without complex conjugate
const int cols = dst_cols/2 + 1;
#else
const int cols = dst_cols;
#endif
__global float2* dst = (__global float2*)(dst_ptr + mad24(y, dst_step, dst_offset));
#pragma unroll
for (int i=x; i<cols; i+=block_size)
dst[i] = SCALE_VAL(smem[i], scale);
#else
// pack row to CCS
__local float* smem_1cn = (__local float*) smem;
__global float* dst = (__global float*)(dst_ptr + mad24(y, dst_step, dst_offset));
for (int i=x; i<dst_cols-1; i+=block_size)
dst[i+1] = SCALE_VAL(smem_1cn[i+2], scale);
if (x == 0)
dst[0] = SCALE_VAL(smem_1cn[0], scale);
#endif
}
else
{
// fill with zero other rows
#ifdef COMPLEX_OUTPUT
__global float2* dst = (__global float2*)(dst_ptr + mad24(y, dst_step, dst_offset));
#else
__global float* dst = (__global float*)(dst_ptr + mad24(y, dst_step, dst_offset));
#endif
#pragma unroll
for (int i=x; i<dst_cols; i+=block_size)
dst[i] = 0.f;
}
}
__kernel void fft_multi_radix_cols(__global const uchar* src_ptr, int src_step, int src_offset, int src_rows, int src_cols,
__global uchar* dst_ptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
__global float2* twiddles_ptr, const int t, const int nz)
{
const int x = get_group_id(0);
const int y = get_global_id(1);
if (x < nz)
{
__local float2 smem[LOCAL_SIZE];
__global const uchar* src = src_ptr + mad24(y, src_step, mad24(x, (int)(sizeof(float)*2), src_offset));
__global const float2* twiddles = (__global float2*) twiddles_ptr;
const int ind = y;
const int block_size = LOCAL_SIZE/kercn;
float scale = 1.f/(dst_rows*dst_cols);
#pragma unroll
for (int i=0; i<kercn; i++)
smem[y+i*block_size] = *((__global const float2*)(src + i*block_size*src_step));
barrier(CLK_LOCAL_MEM_FENCE);
RADIX_PROCESS;
#ifdef COMPLEX_OUTPUT
__global uchar* dst = dst_ptr + mad24(y, dst_step, mad24(x, (int)(sizeof(float)*2), dst_offset));
#pragma unroll
for (int i=0; i<kercn; i++)
*((__global float2*)(dst + i*block_size*dst_step)) = SCALE_VAL(smem[y + i*block_size], scale);
#else
if (x == 0)
{
// pack first column to CCS
__local float* smem_1cn = (__local float*) smem;
__global uchar* dst = dst_ptr + mad24(y+1, dst_step, dst_offset);
for (int i=y; i<dst_rows-1; i+=block_size, dst+=dst_step*block_size)
*((__global float*) dst) = SCALE_VAL(smem_1cn[i+2], scale);
if (y == 0)
*((__global float*) (dst_ptr + dst_offset)) = SCALE_VAL(smem_1cn[0], scale);
}
else if (x == (dst_cols+1)/2)
{
// pack last column to CCS (if needed)
__local float* smem_1cn = (__local float*) smem;
__global uchar* dst = dst_ptr + mad24(dst_cols-1, (int)sizeof(float), mad24(y+1, dst_step, dst_offset));
for (int i=y; i<dst_rows-1; i+=block_size, dst+=dst_step*block_size)
*((__global float*) dst) = SCALE_VAL(smem_1cn[i+2], scale);
if (y == 0)
*((__global float*) (dst_ptr + mad24(dst_cols-1, (int)sizeof(float), dst_offset))) = SCALE_VAL(smem_1cn[0], scale);
}
else
{
__global uchar* dst = dst_ptr + mad24(x, (int)sizeof(float)*2, mad24(y, dst_step, dst_offset - (int)sizeof(float)));
#pragma unroll
for (int i=y; i<dst_rows; i+=block_size, dst+=block_size*dst_step)
vstore2(SCALE_VAL(smem[i], scale), 0, (__global float*) dst);
}
#endif
}
}
__kernel void ifft_multi_radix_rows(__global const uchar* src_ptr, int src_step, int src_offset, int src_rows, int src_cols,
__global uchar* dst_ptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
__global float2* twiddles_ptr, const int t, const int nz)
{
const int x = get_global_id(0);
const int y = get_group_id(1);
const int block_size = LOCAL_SIZE/kercn;
#ifdef IS_1D
const float scale = 1.f/dst_cols;
#else
const float scale = 1.f/(dst_cols*dst_rows);
#endif
if (y < nz)
{
__local float2 smem[LOCAL_SIZE];
__global const float2* twiddles = (__global float2*) twiddles_ptr;
const int ind = x;
#if defined(COMPLEX_INPUT) && !defined(NO_CONJUGATE)
__global const float2* src = (__global const float2*)(src_ptr + mad24(y, src_step, mad24(x, (int)(sizeof(float)*2), src_offset)));
#pragma unroll
for (int i=0; i<kercn; i++)
{
smem[x+i*block_size].x = src[i*block_size].x;
smem[x+i*block_size].y = -src[i*block_size].y;
}
#else
#if !defined(REAL_INPUT) && defined(NO_CONJUGATE)
__global const float2* src = (__global const float2*)(src_ptr + mad24(y, src_step, mad24(2, (int)sizeof(float), src_offset)));
#pragma unroll
for (int i=x; i<(LOCAL_SIZE-1)/2; i+=block_size)
{
smem[i+1].x = src[i].x;
smem[i+1].y = -src[i].y;
smem[LOCAL_SIZE-i-1] = src[i];
}
#else
#pragma unroll
for (int i=x; i<(LOCAL_SIZE-1)/2; i+=block_size)
{
float2 src = vload2(0, (__global const float*)(src_ptr + mad24(y, src_step, mad24(2*i+1, (int)sizeof(float), src_offset))));
smem[i+1].x = src.x;
smem[i+1].y = -src.y;
smem[LOCAL_SIZE-i-1] = src;
}
#endif
if (x==0)
{
smem[0].x = *(__global const float*)(src_ptr + mad24(y, src_step, src_offset));
smem[0].y = 0.f;
if(LOCAL_SIZE % 2 ==0)
{
#if !defined(REAL_INPUT) && defined(NO_CONJUGATE)
smem[LOCAL_SIZE/2].x = src[LOCAL_SIZE/2-1].x;
#else
smem[LOCAL_SIZE/2].x = *(__global const float*)(src_ptr + mad24(y, src_step, mad24(LOCAL_SIZE-1, (int)sizeof(float), src_offset)));
#endif
smem[LOCAL_SIZE/2].y = 0.f;
}
}
#endif
barrier(CLK_LOCAL_MEM_FENCE);
RADIX_PROCESS;
// copy data to dst
#ifdef COMPLEX_OUTPUT
__global float2* dst = (__global float*)(dst_ptr + mad24(y, dst_step, mad24(x, (int)(sizeof(float)*2), dst_offset)));
#pragma unroll
for (int i=0; i<kercn; i++)
{
dst[i*block_size].x = SCALE_VAL(smem[x + i*block_size].x, scale);
dst[i*block_size].y = SCALE_VAL(-smem[x + i*block_size].y, scale);
}
#else
__global float* dst = (__global float*)(dst_ptr + mad24(y, dst_step, mad24(x, (int)(sizeof(float)), dst_offset)));
#pragma unroll
for (int i=0; i<kercn; i++)
{
dst[i*block_size] = SCALE_VAL(smem[x + i*block_size].x, scale);
}
#endif
}
else
{
// fill with zero other rows
#ifdef COMPLEX_OUTPUT
__global float2* dst = (__global float2*)(dst_ptr + mad24(y, dst_step, dst_offset));
#else
__global float* dst = (__global float*)(dst_ptr + mad24(y, dst_step, dst_offset));
#endif
#pragma unroll
for (int i=x; i<dst_cols; i+=block_size)
dst[i] = 0.f;
}
}
__kernel void ifft_multi_radix_cols(__global const uchar* src_ptr, int src_step, int src_offset, int src_rows, int src_cols,
__global uchar* dst_ptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
__global float2* twiddles_ptr, const int t, const int nz)
{
const int x = get_group_id(0);
const int y = get_global_id(1);
#ifdef COMPLEX_INPUT
if (x < nz)
{
__local float2 smem[LOCAL_SIZE];
__global const uchar* src = src_ptr + mad24(y, src_step, mad24(x, (int)(sizeof(float)*2), src_offset));
__global uchar* dst = dst_ptr + mad24(y, dst_step, mad24(x, (int)(sizeof(float)*2), dst_offset));
__global const float2* twiddles = (__global float2*) twiddles_ptr;
const int ind = y;
const int block_size = LOCAL_SIZE/kercn;
#pragma unroll
for (int i=0; i<kercn; i++)
{
float2 temp = *((__global const float2*)(src + i*block_size*src_step));
smem[y+i*block_size].x = temp.x;
smem[y+i*block_size].y = -temp.y;
}
barrier(CLK_LOCAL_MEM_FENCE);
RADIX_PROCESS;
// copy data to dst
#pragma unroll
for (int i=0; i<kercn; i++)
{
__global float2* res = (__global float2*)(dst + i*block_size*dst_step);
res[0].x = smem[y + i*block_size].x;
res[0].y = -smem[y + i*block_size].y;
}
}
#else
if (x < nz)
{
__global const float2* twiddles = (__global float2*) twiddles_ptr;
const int ind = y;
const int block_size = LOCAL_SIZE/kercn;
__local float2 smem[LOCAL_SIZE];
#ifdef EVEN
if (x!=0 && (x!=(nz-1)))
#else
if (x!=0)
#endif
{
__global const uchar* src = src_ptr + mad24(y, src_step, mad24(2*x-1, (int)sizeof(float), src_offset));
#pragma unroll
for (int i=0; i<kercn; i++)
{
float2 temp = vload2(0, (__global const float*)(src + i*block_size*src_step));
smem[y+i*block_size].x = temp.x;
smem[y+i*block_size].y = -temp.y;
}
}
else
{
int ind = x==0 ? 0: 2*x-1;
__global const float* src = (__global const float*)(src_ptr + mad24(1, src_step, mad24(ind, (int)sizeof(float), src_offset)));
int step = src_step/(int)sizeof(float);
#pragma unroll
for (int i=y; i<(LOCAL_SIZE-1)/2; i+=block_size)
{
smem[i+1].x = src[2*i*step];
smem[i+1].y = -src[(2*i+1)*step];
smem[LOCAL_SIZE-i-1].x = src[2*i*step];;
smem[LOCAL_SIZE-i-1].y = src[(2*i+1)*step];
}
if (y==0)
{
smem[0].x = *(__global const float*)(src_ptr + mad24(ind, (int)sizeof(float), src_offset));
smem[0].y = 0.f;
if(LOCAL_SIZE % 2 ==0)
{
smem[LOCAL_SIZE/2].x = src[(LOCAL_SIZE-2)*step];
smem[LOCAL_SIZE/2].y = 0.f;
}
}
}
barrier(CLK_LOCAL_MEM_FENCE);
RADIX_PROCESS;
// copy data to dst
__global uchar* dst = dst_ptr + mad24(y, dst_step, mad24(x, (int)(sizeof(float2)), dst_offset));
#pragma unroll
for (int i=0; i<kercn; i++)
{
__global float2* res = (__global float2*)(dst + i*block_size*dst_step);
res[0].x = smem[y + i*block_size].x;
res[0].y = -smem[y + i*block_size].y;
}
}
#endif
}

2
modules/core/src/opencl/reduce.cl
View File

@ -379,7 +379,7 @@
#define REDUCE_GLOBAL \
dstTK temp = convertToDT(loadpix(srcptr + src_index)); \
dstTK temp2 = convertToDT(loadpix(src2ptr + src2_index)); \
temp = SUM_ABS2(temp, temp2)); \
temp = SUM_ABS2(temp, temp2); \
FUNC(accumulator, temp.s0); \
FUNC(accumulator, temp.s1); \
FUNC(accumulator, temp.s2); \

13
modules/core/src/stat.cpp
View File

@ -479,9 +479,10 @@ static bool ocl_sum( InputArray _src, Scalar & res, int sum_op, InputArray _mask
haveMask = _mask.kind() != _InputArray::NONE,
haveSrc2 = _src2.kind() != _InputArray::NONE;
int type = _src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
kercn = cn == 1 && !haveMask ? ocl::predictOptimalVectorWidth(_src) : 1,
kercn = cn == 1 && !haveMask ? ocl::predictOptimalVectorWidth(_src, _src2) : 1,
mcn = std::max(cn, kercn);
CV_Assert(!haveSrc2 || _src2.type() == type);
int convert_cn = haveSrc2 ? mcn : cn;
if ( (!doubleSupport && depth == CV_64F) || cn > 4 )
return false;
@ -513,7 +514,7 @@ static bool ocl_sum( InputArray _src, Scalar & res, int sum_op, InputArray _mask
haveMask && _mask.isContinuous() ? " -D HAVE_MASK_CONT" : "", kercn,
haveSrc2 ? " -D HAVE_SRC2" : "", calc2 ? " -D OP_CALC2" : "",
haveSrc2 && _src2.isContinuous() ? " -D HAVE_SRC2_CONT" : "",
depth <= CV_32S && ddepth == CV_32S ? ocl::convertTypeStr(CV_8U, ddepth, mcn, cvt[1]) : "noconvert");
depth <= CV_32S && ddepth == CV_32S ? ocl::convertTypeStr(CV_8U, ddepth, convert_cn, cvt[1]) : "noconvert");
ocl::Kernel k("reduce", ocl::core::reduce_oclsrc, opts);
if (k.empty())
@ -1451,6 +1452,9 @@ static bool ocl_minMaxIdx( InputArray _src, double* minVal, double* maxVal, int*
CV_Assert(!haveSrc2 || _src2.type() == type);
if (depth == CV_32S || depth == CV_32F)
return false;
if ((depth == CV_64F || ddepth == CV_64F) && !doubleSupport)
return false;
@ -2184,6 +2188,9 @@ static bool ocl_norm( InputArray _src, int normType, InputArray _mask, double &
(!doubleSupport && depth == CV_64F))
return false;
if( depth == CV_32F && (!_mask.empty() || normType == NORM_INF) )
return false;
UMat src = _src.getUMat();
if (normType == NORM_INF)
@ -2539,7 +2546,7 @@ static bool ocl_norm( InputArray _src1, InputArray _src2, int normType, InputArr
normType &= ~NORM_RELATIVE;
bool normsum = normType == NORM_L1 || normType == NORM_L2 || normType == NORM_L2SQR;
if ( !(normType == NORM_INF || normsum) )
if ( !normsum || !_mask.empty() )
return false;
if (normsum)

65
modules/core/test/ocl/test_dft.cpp
View File

@ -48,17 +48,26 @@
#ifdef HAVE_OPENCL
enum OCL_FFT_TYPE
{
R2R = 0,
C2R = 1,
R2C = 2,
C2C = 3
};
namespace cvtest {
namespace ocl {
////////////////////////////////////////////////////////////////////////////
// Dft
PARAM_TEST_CASE(Dft, cv::Size, MatDepth, bool, bool, bool, bool)
PARAM_TEST_CASE(Dft, cv::Size, OCL_FFT_TYPE, bool, bool, bool, bool)
{
cv::Size dft_size;
int dft_flags, depth;
bool inplace;
int dft_flags, depth, cn, dft_type;
bool hint;
bool is1d;
TEST_DECLARE_INPUT_PARAMETER(src);
TEST_DECLARE_OUTPUT_PARAMETER(dst);
@ -66,34 +75,50 @@ PARAM_TEST_CASE(Dft, cv::Size, MatDepth, bool, bool, bool, bool)
virtual void SetUp()
{
dft_size = GET_PARAM(0);
depth = GET_PARAM(1);
inplace = GET_PARAM(2);
dft_type = GET_PARAM(1);
depth = CV_32F;
dft_flags = 0;
switch (dft_type)
{
case R2R: dft_flags |= cv::DFT_REAL_OUTPUT; cn = 1; break;
case C2R: dft_flags |= cv::DFT_REAL_OUTPUT; cn = 2; break;
case R2C: dft_flags |= cv::DFT_COMPLEX_OUTPUT; cn = 1; break;
case C2C: dft_flags |= cv::DFT_COMPLEX_OUTPUT; cn = 2; break;
}
if (GET_PARAM(2))
dft_flags |= cv::DFT_INVERSE;
if (GET_PARAM(3))
dft_flags |= cv::DFT_ROWS;
if (GET_PARAM(4))
dft_flags |= cv::DFT_SCALE;
if (GET_PARAM(5))
dft_flags |= cv::DFT_INVERSE;
hint = GET_PARAM(5);
is1d = (dft_flags & DFT_ROWS) != 0 || dft_size.height == 1;
}
void generateTestData(int cn = 2)
void generateTestData()
{
src = randomMat(dft_size, CV_MAKE_TYPE(depth, cn), 0.0, 100.0);
usrc = src.getUMat(ACCESS_READ);
if (inplace)
dst = src, udst = usrc;
}
};
OCL_TEST_P(Dft, C2C)
OCL_TEST_P(Dft, Mat)
{
generateTestData();
OCL_OFF(cv::dft(src, dst, dft_flags | cv::DFT_COMPLEX_OUTPUT));
OCL_ON(cv::dft(usrc, udst, dft_flags | cv::DFT_COMPLEX_OUTPUT));
int nonzero_rows = hint ? src.cols - randomInt(1, src.rows-1) : 0;
OCL_OFF(cv::dft(src, dst, dft_flags, nonzero_rows));
OCL_ON(cv::dft(usrc, udst, dft_flags, nonzero_rows));
// In case forward R2C 1d tranform dst contains only half of output
// without complex conjugate
if (dft_type == R2C && is1d && (dft_flags & cv::DFT_INVERSE) == 0)
{
dst = dst(cv::Range(0, dst.rows), cv::Range(0, dst.cols/2 + 1));
udst = udst(cv::Range(0, udst.rows), cv::Range(0, udst.cols/2 + 1));
}
double eps = src.size().area() * 1e-4;
EXPECT_MAT_NEAR(dst, udst, eps);
@ -150,15 +175,15 @@ OCL_TEST_P(MulSpectrums, Mat)
OCL_INSTANTIATE_TEST_CASE_P(OCL_ImgProc, MulSpectrums, testing::Combine(Bool(), Bool()));
OCL_INSTANTIATE_TEST_CASE_P(Core, Dft, Combine(Values(cv::Size(2, 3), cv::Size(5, 4), cv::Size(25, 20),
cv::Size(512, 1), cv::Size(1024, 768)),
Values(CV_32F, CV_64F),
Bool(), // inplace
OCL_INSTANTIATE_TEST_CASE_P(Core, Dft, Combine(Values(cv::Size(10, 10), cv::Size(36, 36), cv::Size(512, 1), cv::Size(1280, 768)),
Values((OCL_FFT_TYPE) R2C, (OCL_FFT_TYPE) C2C, (OCL_FFT_TYPE) R2R, (OCL_FFT_TYPE) C2R),
Bool(), // DFT_INVERSE
Bool(), // DFT_ROWS
Bool(), // DFT_SCALE
Bool()) // DFT_INVERSE
Bool() // hint
)
);
} } // namespace cvtest::ocl
#endif // HAVE_OPENCL
#endif // HAVE_OPENCL

8
modules/cudabgsegm/perf/perf_bgsegm.cpp
View File

@ -42,8 +42,8 @@
#include "perf_precomp.hpp"
#ifdef HAVE_OPENCV_LEGACY
# include "opencv2/legacy.hpp"
#ifdef HAVE_OPENCV_CUDALEGACY
# include "opencv2/cudalegacy.hpp"
#endif
#ifdef HAVE_OPENCV_CUDAIMGPROC
@ -72,7 +72,7 @@ using namespace perf;
#if BUILD_WITH_VIDEO_INPUT_SUPPORT
#ifdef HAVE_OPENCV_LEGACY
#ifdef HAVE_OPENCV_CUDALEGACY
namespace cv
{
@ -150,7 +150,7 @@ PERF_TEST_P(Video, FGDStatModel,
}
else
{
#ifdef HAVE_OPENCV_LEGACY
#ifdef HAVE_OPENCV_CUDALEGACY
IplImage ipl_frame = frame;
cv::Ptr<CvBGStatModel> model(cvCreateFGDStatModel(&ipl_frame));

6
modules/cudabgsegm/test/test_bgsegm.cpp
View File

@ -42,8 +42,8 @@
#include "test_precomp.hpp"
#ifdef HAVE_OPENCV_LEGACY
# include "opencv2/legacy.hpp"
#ifdef HAVE_OPENCV_CUDALEGACY
# include "opencv2/cudalegacy.hpp"
#endif
#ifdef HAVE_CUDA
@ -66,7 +66,7 @@ using namespace cvtest;
//////////////////////////////////////////////////////
// FGDStatModel
#if BUILD_WITH_VIDEO_INPUT_SUPPORT && defined(HAVE_OPENCV_LEGACY)
#if BUILD_WITH_VIDEO_INPUT_SUPPORT && defined(HAVE_OPENCV_CUDALEGACY)
namespace cv
{

2
modules/cudacodec/CMakeLists.txt
View File

@ -6,7 +6,7 @@ set(the_description "CUDA-accelerated Video Encoding/Decoding")
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4127 /wd4324 /wd4512 -Wundef)
ocv_add_module(cudacodec OPTIONAL opencv_cudev)
ocv_add_module(cudacodec opencv_core opencv_videoio OPTIONAL opencv_cudev)
ocv_module_include_directories()
ocv_glob_module_sources()

46
modules/cudastereo/src/cuda/disparity_bilateral_filter.cu
View File

@ -45,34 +45,12 @@
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/limits.hpp"
#include "cuda/disparity_bilateral_filter.hpp"
namespace cv { namespace cuda { namespace device
{
namespace disp_bilateral_filter
{
__constant__ float* ctable_color;
__constant__ float* ctable_space;
__constant__ size_t ctable_space_step;
__constant__ int cndisp;
__constant__ int cradius;
__constant__ short cedge_disc;
__constant__ short cmax_disc;
void disp_load_constants(float* table_color, PtrStepSzf table_space, int ndisp, int radius, short edge_disc, short max_disc)
{
cudaSafeCall( cudaMemcpyToSymbol(ctable_color, &table_color, sizeof(table_color)) );
cudaSafeCall( cudaMemcpyToSymbol(ctable_space, &table_space.data, sizeof(table_space.data)) );
size_t table_space_step = table_space.step / sizeof(float);
cudaSafeCall( cudaMemcpyToSymbol(ctable_space_step, &table_space_step, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cndisp, &ndisp, sizeof(int)) );
cudaSafeCall( cudaMemcpyToSymbol(cradius, &radius, sizeof(int)) );
cudaSafeCall( cudaMemcpyToSymbol(cedge_disc, &edge_disc, sizeof(short)) );
cudaSafeCall( cudaMemcpyToSymbol(cmax_disc, &max_disc, sizeof(short)) );
}
template <int channels>
struct DistRgbMax
{
@ -95,7 +73,11 @@ namespace cv { namespace cuda { namespace device
};
template <int channels, typename T>
__global__ void disp_bilateral_filter(int t, T* disp, size_t disp_step, const uchar* img, size_t img_step, int h, int w)
__global__ void disp_bilateral_filter(int t, T* disp, size_t disp_step,
const uchar* img, size_t img_step, int h, int w,
const float* ctable_color, const float * ctable_space, size_t ctable_space_step,
int cradius,
short cedge_disc, short cmax_disc)
{
const int y = blockIdx.y * blockDim.y + threadIdx.y;
const int x = ((blockIdx.x * blockDim.x + threadIdx.x) << 1) + ((y + t) & 1);
@ -178,7 +160,7 @@ namespace cv { namespace cuda { namespace device
}
template <typename T>
void disp_bilateral_filter(PtrStepSz<T> disp, PtrStepSzb img, int channels, int iters, cudaStream_t stream)
void disp_bilateral_filter(PtrStepSz<T> disp, PtrStepSzb img, int channels, int iters, const float *table_color, const float* table_space, size_t table_step, int radius, short edge_disc, short max_disc, cudaStream_t stream)
{
dim3 threads(32, 8, 1);
dim3 grid(1, 1, 1);
@ -190,20 +172,20 @@ namespace cv { namespace cuda { namespace device
case 1:
for (int i = 0; i < iters; ++i)
{
disp_bilateral_filter<1><<<grid, threads, 0, stream>>>(0, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols);
disp_bilateral_filter<1><<<grid, threads, 0, stream>>>(0, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols, table_color, table_space, table_step, radius, edge_disc, max_disc);
cudaSafeCall( cudaGetLastError() );
disp_bilateral_filter<1><<<grid, threads, 0, stream>>>(1, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols);
disp_bilateral_filter<1><<<grid, threads, 0, stream>>>(1, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols, table_color, table_space, table_step, radius, edge_disc, max_disc);
cudaSafeCall( cudaGetLastError() );
}
break;
case 3:
for (int i = 0; i < iters; ++i)
{
disp_bilateral_filter<3><<<grid, threads, 0, stream>>>(0, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols);
disp_bilateral_filter<3><<<grid, threads, 0, stream>>>(0, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols, table_color, table_space, table_step, radius, edge_disc, max_disc);
cudaSafeCall( cudaGetLastError() );
disp_bilateral_filter<3><<<grid, threads, 0, stream>>>(1, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols);
disp_bilateral_filter<3><<<grid, threads, 0, stream>>>(1, disp.data, disp.step/sizeof(T), img.data, img.step, disp.rows, disp.cols, table_color, table_space, table_step, radius, edge_disc, max_disc);
cudaSafeCall( cudaGetLastError() );
}
break;
@ -215,8 +197,8 @@ namespace cv { namespace cuda { namespace device
cudaSafeCall( cudaDeviceSynchronize() );
}
template void disp_bilateral_filter<uchar>(PtrStepSz<uchar> disp, PtrStepSzb img, int channels, int iters, cudaStream_t stream);
template void disp_bilateral_filter<short>(PtrStepSz<short> disp, PtrStepSzb img, int channels, int iters, cudaStream_t stream);
template void disp_bilateral_filter<uchar>(PtrStepSz<uchar> disp, PtrStepSzb img, int channels, int iters, const float *table_color, const float *table_space, size_t table_step, int radius, short, short, cudaStream_t stream);
template void disp_bilateral_filter<short>(PtrStepSz<short> disp, PtrStepSzb img, int channels, int iters, const float *table_color, const float *table_space, size_t table_step, int radius, short, short, cudaStream_t stream);
} // namespace bilateral_filter
}}} // namespace cv { namespace cuda { namespace cudev

8
modules/cudastereo/src/cuda/disparity_bilateral_filter.hpp Normal file
View File

@ -0,0 +1,8 @@
namespace cv { namespace cuda { namespace device
{
namespace disp_bilateral_filter
{
template<typename T>
void disp_bilateral_filter(PtrStepSz<T> disp, PtrStepSzb img, int channels, int iters, const float *, const float *, size_t, int radius, short edge_disc, short max_disc, cudaStream_t stream);
}
}}}

442
modules/cudastereo/src/cuda/stereocsbp.cu
View File

@ -48,109 +48,61 @@
#include "opencv2/core/cuda/reduce.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "cuda/stereocsbp.hpp"
namespace cv { namespace cuda { namespace device
{
namespace stereocsbp
{
///////////////////////////////////////////////////////////////
/////////////////////// load constants ////////////////////////
///////////////////////////////////////////////////////////////
__constant__ int cndisp;
__constant__ float cmax_data_term;
__constant__ float cdata_weight;
__constant__ float cmax_disc_term;
__constant__ float cdisc_single_jump;
__constant__ int cth;
__constant__ size_t cimg_step;
__constant__ size_t cmsg_step;
__constant__ size_t cdisp_step1;
__constant__ size_t cdisp_step2;
__constant__ uchar* cleft;
__constant__ uchar* cright;
__constant__ uchar* ctemp;
void load_constants(int ndisp, float max_data_term, float data_weight, float max_disc_term, float disc_single_jump, int min_disp_th,
const PtrStepSzb& left, const PtrStepSzb& right, const PtrStepSzb& temp)
{
cudaSafeCall( cudaMemcpyToSymbol(cndisp, &ndisp, sizeof(int)) );
cudaSafeCall( cudaMemcpyToSymbol(cmax_data_term, &max_data_term, sizeof(float)) );
cudaSafeCall( cudaMemcpyToSymbol(cdata_weight, &data_weight, sizeof(float)) );
cudaSafeCall( cudaMemcpyToSymbol(cmax_disc_term, &max_disc_term, sizeof(float)) );
cudaSafeCall( cudaMemcpyToSymbol(cdisc_single_jump, &disc_single_jump, sizeof(float)) );
cudaSafeCall( cudaMemcpyToSymbol(cth, &min_disp_th, sizeof(int)) );
cudaSafeCall( cudaMemcpyToSymbol(cimg_step, &left.step, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cleft, &left.data, sizeof(left.data)) );
cudaSafeCall( cudaMemcpyToSymbol(cright, &right.data, sizeof(right.data)) );
cudaSafeCall( cudaMemcpyToSymbol(ctemp, &temp.data, sizeof(temp.data)) );
}
///////////////////////////////////////////////////////////////
/////////////////////// init data cost ////////////////////////
///////////////////////////////////////////////////////////////
template <int channels> struct DataCostPerPixel;
template <> struct DataCostPerPixel<1>
template <int channels> static float __device__ pixeldiff(const uchar* left, const uchar* right, float max_data_term);
template<> __device__ __forceinline__ static float pixeldiff<1>(const uchar* left, const uchar* right, float max_data_term)
{
static __device__ __forceinline__ float compute(const uchar* left, const uchar* right)
{
return fmin(cdata_weight * ::abs((int)*left - *right), cdata_weight * cmax_data_term);
}
};
template <> struct DataCostPerPixel<3>
return fmin( ::abs((int)*left - *right), max_data_term);
}
template<> __device__ __forceinline__ static float pixeldiff<3>(const uchar* left, const uchar* right, float max_data_term)
{
static __device__ __forceinline__ float compute(const uchar* left, const uchar* right)
{
float tb = 0.114f * ::abs((int)left[0] - right[0]);
float tg = 0.587f * ::abs((int)left[1] - right[1]);
float tr = 0.299f * ::abs((int)left[2] - right[2]);
float tb = 0.114f * ::abs((int)left[0] - right[0]);
float tg = 0.587f * ::abs((int)left[1] - right[1]);
float tr = 0.299f * ::abs((int)left[2] - right[2]);
return fmin(cdata_weight * (tr + tg + tb), cdata_weight * cmax_data_term);
}
};
template <> struct DataCostPerPixel<4>
return fmin(tr + tg + tb, max_data_term);
}
template<> __device__ __forceinline__ static float pixeldiff<4>(const uchar* left, const uchar* right, float max_data_term)
{
static __device__ __forceinline__ float compute(const uchar* left, const uchar* right)
{
uchar4 l = *((const uchar4*)left);
uchar4 r = *((const uchar4*)right);
uchar4 l = *((const uchar4*)left);
uchar4 r = *((const uchar4*)right);
float tb = 0.114f * ::abs((int)l.x - r.x);
float tg = 0.587f * ::abs((int)l.y - r.y);
float tr = 0.299f * ::abs((int)l.z - r.z);
float tb = 0.114f * ::abs((int)l.x - r.x);
float tg = 0.587f * ::abs((int)l.y - r.y);
float tr = 0.299f * ::abs((int)l.z - r.z);
return fmin(cdata_weight * (tr + tg + tb), cdata_weight * cmax_data_term);
}
};
return fmin(tr + tg + tb, max_data_term);
}
template <typename T>
__global__ void get_first_k_initial_global(T* data_cost_selected_, T *selected_disp_pyr, int h, int w, int nr_plane)
__global__ void get_first_k_initial_global(uchar *ctemp, T* data_cost_selected_, T *selected_disp_pyr, int h, int w, int nr_plane, int ndisp,
size_t msg_step, size_t disp_step)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y < h && x < w)
{
T* selected_disparity = selected_disp_pyr + y * cmsg_step + x;
T* data_cost_selected = data_cost_selected_ + y * cmsg_step + x;
T* data_cost = (T*)ctemp + y * cmsg_step + x;
T* selected_disparity = selected_disp_pyr + y * msg_step + x;
T* data_cost_selected = data_cost_selected_ + y * msg_step + x;
T* data_cost = (T*)ctemp + y * msg_step + x;
for(int i = 0; i < nr_plane; i++)
{
T minimum = device::numeric_limits<T>::max();
int id = 0;
for(int d = 0; d < cndisp; d++)
for(int d = 0; d < ndisp; d++)
{
T cur = data_cost[d * cdisp_step1];
T cur = data_cost[d * disp_step];
if(cur < minimum)
{
minimum = cur;
@ -158,46 +110,47 @@ namespace cv { namespace cuda { namespace device
}
}
data_cost_selected[i * cdisp_step1] = minimum;
selected_disparity[i * cdisp_step1] = id;
data_cost [id * cdisp_step1] = numeric_limits<T>::max();
data_cost_selected[i * disp_step] = minimum;
selected_disparity[i * disp_step] = id;
data_cost [id * disp_step] = numeric_limits<T>::max();
}
}
}
template <typename T>
__global__ void get_first_k_initial_local(T* data_cost_selected_, T* selected_disp_pyr, int h, int w, int nr_plane)
__global__ void get_first_k_initial_local(uchar *ctemp, T* data_cost_selected_, T* selected_disp_pyr, int h, int w, int nr_plane, int ndisp,
size_t msg_step, size_t disp_step)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y < h && x < w)
{
T* selected_disparity = selected_disp_pyr + y * cmsg_step + x;
T* data_cost_selected = data_cost_selected_ + y * cmsg_step + x;
T* data_cost = (T*)ctemp + y * cmsg_step + x;
T* selected_disparity = selected_disp_pyr + y * msg_step + x;
T* data_cost_selected = data_cost_selected_ + y * msg_step + x;
T* data_cost = (T*)ctemp + y * msg_step + x;
int nr_local_minimum = 0;
T prev = data_cost[0 * cdisp_step1];
T cur = data_cost[1 * cdisp_step1];
T next = data_cost[2 * cdisp_step1];
T prev = data_cost[0 * disp_step];
T cur = data_cost[1 * disp_step];
T next = data_cost[2 * disp_step];
for (int d = 1; d < cndisp - 1 && nr_local_minimum < nr_plane; d++)
for (int d = 1; d < ndisp - 1 && nr_local_minimum < nr_plane; d++)
{
if (cur < prev && cur < next)
{
data_cost_selected[nr_local_minimum * cdisp_step1] = cur;
selected_disparity[nr_local_minimum * cdisp_step1] = d;
data_cost_selected[nr_local_minimum * disp_step] = cur;
selected_disparity[nr_local_minimum * disp_step] = d;
data_cost[d * cdisp_step1] = numeric_limits<T>::max();
data_cost[d * disp_step] = numeric_limits<T>::max();
nr_local_minimum++;
}
prev = cur;
cur = next;
next = data_cost[(d + 1) * cdisp_step1];
next = data_cost[(d + 1) * disp_step];
}
for (int i = nr_local_minimum; i < nr_plane; i++)
@ -205,25 +158,27 @@ namespace cv { namespace cuda { namespace device
T minimum = numeric_limits<T>::max();
int id = 0;
for (int d = 0; d < cndisp; d++)
for (int d = 0; d < ndisp; d++)
{
cur = data_cost[d * cdisp_step1];
cur = data_cost[d * disp_step];
if (cur < minimum)
{
minimum = cur;
id = d;
}
}
data_cost_selected[i * cdisp_step1] = minimum;
selected_disparity[i * cdisp_step1] = id;
data_cost_selected[i * disp_step] = minimum;
selected_disparity[i * disp_step] = id;
data_cost[id * cdisp_step1] = numeric_limits<T>::max();
data_cost[id * disp_step] = numeric_limits<T>::max();
}
}
}
template <typename T, int channels>
__global__ void init_data_cost(int h, int w, int level)
__global__ void init_data_cost(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step,
int h, int w, int level, int ndisp, float data_weight, float max_data_term,
int min_disp, size_t msg_step, size_t disp_step)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
@ -236,9 +191,9 @@ namespace cv { namespace cuda { namespace device
int x0 = x << level;
int xt = (x + 1) << level;
T* data_cost = (T*)ctemp + y * cmsg_step + x;
T* data_cost = (T*)ctemp + y * msg_step + x;
for(int d = 0; d < cndisp; ++d)
for(int d = 0; d < ndisp; ++d)
{
float val = 0.0f;
for(int yi = y0; yi < yt; yi++)
@ -246,24 +201,26 @@ namespace cv { namespace cuda { namespace device
for(int xi = x0; xi < xt; xi++)
{
int xr = xi - d;
if(d < cth || xr < 0)
val += cdata_weight * cmax_data_term;
if(d < min_disp || xr < 0)
val += data_weight * max_data_term;
else
{
const uchar* lle = cleft + yi * cimg_step + xi * channels;
const uchar* lri = cright + yi * cimg_step + xr * channels;
val += DataCostPerPixel<channels>::compute(lle, lri);
val += data_weight * pixeldiff<channels>(lle, lri, max_data_term);
}
}
}
data_cost[cdisp_step1 * d] = saturate_cast<T>(val);
data_cost[disp_step * d] = saturate_cast<T>(val);
}
}
}
template <typename T, int winsz, int channels>
__global__ void init_data_cost_reduce(int level, int rows, int cols, int h)
__global__ void init_data_cost_reduce(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step,
int level, int rows, int cols, int h, int ndisp, float data_weight, float max_data_term,
int min_disp, size_t msg_step, size_t disp_step)
{
int x_out = blockIdx.x;
int y_out = blockIdx.y % h;
@ -271,7 +228,7 @@ namespace cv { namespace cuda { namespace device
int tid = threadIdx.x;
if (d < cndisp)
if (d < ndisp)
{
int x0 = x_out << level;
int y0 = y_out << level;
@ -281,8 +238,8 @@ namespace cv { namespace cuda { namespace device
float val = 0.0f;
if (x0 + tid < cols)
{
if (x0 + tid - d < 0 || d < cth)
val = cdata_weight * cmax_data_term * len;
if (x0 + tid - d < 0 || d < min_disp)
val = data_weight * max_data_term * len;
else
{
const uchar* lle = cleft + y0 * cimg_step + channels * (x0 + tid );
@ -290,7 +247,7 @@ namespace cv { namespace cuda { namespace device
for(int y = 0; y < len; ++y)
{
val += DataCostPerPixel<channels>::compute(lle, lri);
val += data_weight * pixeldiff<channels>(lle, lri, max_data_term);
lle += cimg_step;
lri += cimg_step;
@ -302,16 +259,16 @@ namespace cv { namespace cuda { namespace device
reduce<winsz>(smem + winsz * threadIdx.z, val, tid, plus<float>());
T* data_cost = (T*)ctemp + y_out * cmsg_step + x_out;
T* data_cost = (T*)ctemp + y_out * msg_step + x_out;
if (tid == 0)
data_cost[cdisp_step1 * d] = saturate_cast<T>(val);
data_cost[disp_step * d] = saturate_cast<T>(val);
}
}
template <typename T>
void init_data_cost_caller_(int /*rows*/, int /*cols*/, int h, int w, int level, int /*ndisp*/, int channels, cudaStream_t stream)
void init_data_cost_caller_(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step, int /*rows*/, int /*cols*/, int h, int w, int level, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step, cudaStream_t stream)
{
dim3 threads(32, 8, 1);
dim3 grid(1, 1, 1);
@ -321,15 +278,15 @@ namespace cv { namespace cuda { namespace device
switch (channels)
{
case 1: init_data_cost<T, 1><<<grid, threads, 0, stream>>>(h, w, level); break;
case 3: init_data_cost<T, 3><<<grid, threads, 0, stream>>>(h, w, level); break;
case 4: init_data_cost<T, 4><<<grid, threads, 0, stream>>>(h, w, level); break;
case 1: init_data_cost<T, 1><<<grid, threads, 0, stream>>>(cleft, cright, ctemp, cimg_step, h, w, level, ndisp, data_weight, max_data_term, min_disp, msg_step, disp_step); break;
case 3: init_data_cost<T, 3><<<grid, threads, 0, stream>>>(cleft, cright, ctemp, cimg_step, h, w, level, ndisp, data_weight, max_data_term, min_disp, msg_step, disp_step); break;
case 4: init_data_cost<T, 4><<<grid, threads, 0, stream>>>(cleft, cright, ctemp, cimg_step, h, w, level, ndisp, data_weight, max_data_term, min_disp, msg_step, disp_step); break;
default: CV_Error(cv::Error::BadNumChannels, "Unsupported channels count");
}
}
template <typename T, int winsz>
void init_data_cost_reduce_caller_(int rows, int cols, int h, int w, int level, int ndisp, int channels, cudaStream_t stream)
void init_data_cost_reduce_caller_(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step, int rows, int cols, int h, int w, int level, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step, cudaStream_t stream)
{
const int threadsNum = 256;
const size_t smem_size = threadsNum * sizeof(float);
@ -340,19 +297,19 @@ namespace cv { namespace cuda { namespace device
switch (channels)
{
case 1: init_data_cost_reduce<T, winsz, 1><<<grid, threads, smem_size, stream>>>(level, rows, cols, h); break;
case 3: init_data_cost_reduce<T, winsz, 3><<<grid, threads, smem_size, stream>>>(level, rows, cols, h); break;
case 4: init_data_cost_reduce<T, winsz, 4><<<grid, threads, smem_size, stream>>>(level, rows, cols, h); break;
case 1: init_data_cost_reduce<T, winsz, 1><<<grid, threads, smem_size, stream>>>(cleft, cright, ctemp, cimg_step, level, rows, cols, h, ndisp, data_weight, max_data_term, min_disp, msg_step, disp_step); break;
case 3: init_data_cost_reduce<T, winsz, 3><<<grid, threads, smem_size, stream>>>(cleft, cright, ctemp, cimg_step, level, rows, cols, h, ndisp, data_weight, max_data_term, min_disp, msg_step, disp_step); break;
case 4: init_data_cost_reduce<T, winsz, 4><<<grid, threads, smem_size, stream>>>(cleft, cright, ctemp, cimg_step, level, rows, cols, h, ndisp, data_weight, max_data_term, min_disp, msg_step, disp_step); break;
default: CV_Error(cv::Error::BadNumChannels, "Unsupported channels count");
}
}
template<class T>
void init_data_cost(int rows, int cols, T* disp_selected_pyr, T* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, bool use_local_init_data_cost, cudaStream_t stream)
void init_data_cost(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step, int rows, int cols, T* disp_selected_pyr, T* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, bool use_local_init_data_cost, cudaStream_t stream)
{
typedef void (*InitDataCostCaller)(int cols, int rows, int w, int h, int level, int ndisp, int channels, cudaStream_t stream);
typedef void (*InitDataCostCaller)(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step, int cols, int rows, int w, int h, int level, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step, cudaStream_t stream);
static const InitDataCostCaller init_data_cost_callers[] =
{
@ -362,10 +319,8 @@ namespace cv { namespace cuda { namespace device
};
size_t disp_step = msg_step * h;
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step1, &disp_step, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cmsg_step, &msg_step, sizeof(size_t)) );
init_data_cost_callers[level](rows, cols, h, w, level, ndisp, channels, stream);
init_data_cost_callers[level](cleft, cright, ctemp, cimg_step, rows, cols, h, w, level, ndisp, channels, data_weight, max_data_term, min_disp, msg_step, disp_step, stream);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
@ -378,9 +333,9 @@ namespace cv { namespace cuda { namespace device
grid.y = divUp(h, threads.y);
if (use_local_init_data_cost == true)
get_first_k_initial_local<<<grid, threads, 0, stream>>> (data_cost_selected, disp_selected_pyr, h, w, nr_plane);
get_first_k_initial_local<<<grid, threads, 0, stream>>> (ctemp, data_cost_selected, disp_selected_pyr, h, w, nr_plane, ndisp, msg_step, disp_step);
else
get_first_k_initial_global<<<grid, threads, 0, stream>>>(data_cost_selected, disp_selected_pyr, h, w, nr_plane);
get_first_k_initial_global<<<grid, threads, 0, stream>>>(ctemp, data_cost_selected, disp_selected_pyr, h, w, nr_plane, ndisp, msg_step, disp_step);
cudaSafeCall( cudaGetLastError() );
@ -388,18 +343,18 @@ namespace cv { namespace cuda { namespace device
cudaSafeCall( cudaDeviceSynchronize() );
}
template void init_data_cost(int rows, int cols, short* disp_selected_pyr, short* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, bool use_local_init_data_cost, cudaStream_t stream);
template void init_data_cost<short>(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step, int rows, int cols, short* disp_selected_pyr, short* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, bool use_local_init_data_cost, cudaStream_t stream);
template void init_data_cost(int rows, int cols, float* disp_selected_pyr, float* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, bool use_local_init_data_cost, cudaStream_t stream);
template void init_data_cost<float>(const uchar *cleft, const uchar *cright, uchar *ctemp, size_t cimg_step, int rows, int cols, float* disp_selected_pyr, float* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, bool use_local_init_data_cost, cudaStream_t stream);
///////////////////////////////////////////////////////////////
////////////////////// compute data cost //////////////////////
///////////////////////////////////////////////////////////////
template <typename T, int channels>
__global__ void compute_data_cost(const T* selected_disp_pyr, T* data_cost_, int h, int w, int level, int nr_plane)
__global__ void compute_data_cost(const uchar *cleft, const uchar *cright, size_t cimg_step, const T* selected_disp_pyr, T* data_cost_, int h, int w, int level, int nr_plane, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step1, size_t disp_step2)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
@ -412,8 +367,8 @@ namespace cv { namespace cuda { namespace device
int x0 = x << level;
int xt = (x + 1) << level;
const T* selected_disparity = selected_disp_pyr + y/2 * cmsg_step + x/2;
T* data_cost = data_cost_ + y * cmsg_step + x;
const T* selected_disparity = selected_disp_pyr + y/2 * msg_step + x/2;
T* data_cost = data_cost_ + y * msg_step + x;
for(int d = 0; d < nr_plane; d++)
{
@ -422,27 +377,27 @@ namespace cv { namespace cuda { namespace device
{
for(int xi = x0; xi < xt; xi++)
{
int sel_disp = selected_disparity[d * cdisp_step2];
int sel_disp = selected_disparity[d * disp_step2];
int xr = xi - sel_disp;
if (xr < 0 || sel_disp < cth)
val += cdata_weight * cmax_data_term;
if (xr < 0 || sel_disp < min_disp)
val += data_weight * max_data_term;
else
{
const uchar* left_x = cleft + yi * cimg_step + xi * channels;
const uchar* right_x = cright + yi * cimg_step + xr * channels;
val += DataCostPerPixel<channels>::compute(left_x, right_x);
val += data_weight * pixeldiff<channels>(left_x, right_x, max_data_term);
}
}
}
data_cost[cdisp_step1 * d] = saturate_cast<T>(val);
data_cost[disp_step1 * d] = saturate_cast<T>(val);
}
}
}
template <typename T, int winsz, int channels>
__global__ void compute_data_cost_reduce(const T* selected_disp_pyr, T* data_cost_, int level, int rows, int cols, int h, int nr_plane)
__global__ void compute_data_cost_reduce(const uchar *cleft, const uchar *cright, size_t cimg_step, const T* selected_disp_pyr, T* data_cost_, int level, int rows, int cols, int h, int nr_plane, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step1, size_t disp_step2)
{
int x_out = blockIdx.x;
int y_out = blockIdx.y % h;
@ -450,12 +405,12 @@ namespace cv { namespace cuda { namespace device
int tid = threadIdx.x;
const T* selected_disparity = selected_disp_pyr + y_out/2 * cmsg_step + x_out/2;
T* data_cost = data_cost_ + y_out * cmsg_step + x_out;
const T* selected_disparity = selected_disp_pyr + y_out/2 * msg_step + x_out/2;
T* data_cost = data_cost_ + y_out * msg_step + x_out;
if (d < nr_plane)
{
int sel_disp = selected_disparity[d * cdisp_step2];
int sel_disp = selected_disparity[d * disp_step2];
int x0 = x_out << level;
int y0 = y_out << level;
@ -465,8 +420,8 @@ namespace cv { namespace cuda { namespace device
float val = 0.0f;
if (x0 + tid < cols)
{
if (x0 + tid - sel_disp < 0 || sel_disp < cth)
val = cdata_weight * cmax_data_term * len;
if (x0 + tid - sel_disp < 0 || sel_disp < min_disp)
val = data_weight * max_data_term * len;
else
{
const uchar* lle = cleft + y0 * cimg_step + channels * (x0 + tid );
@ -474,7 +429,7 @@ namespace cv { namespace cuda { namespace device
for(int y = 0; y < len; ++y)
{
val += DataCostPerPixel<channels>::compute(lle, lri);
val += data_weight * pixeldiff<channels>(lle, lri, max_data_term);
lle += cimg_step;
lri += cimg_step;
@ -487,13 +442,13 @@ namespace cv { namespace cuda { namespace device
reduce<winsz>(smem + winsz * threadIdx.z, val, tid, plus<float>());
if (tid == 0)
data_cost[cdisp_step1 * d] = saturate_cast<T>(val);
data_cost[disp_step1 * d] = saturate_cast<T>(val);
}
}
template <typename T>
void compute_data_cost_caller_(const T* disp_selected_pyr, T* data_cost, int /*rows*/, int /*cols*/,
int h, int w, int level, int nr_plane, int channels, cudaStream_t stream)
void compute_data_cost_caller_(const uchar *cleft, const uchar *cright, size_t cimg_step, const T* disp_selected_pyr, T* data_cost, int /*rows*/, int /*cols*/,
int h, int w, int level, int nr_plane, int channels, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step1, size_t disp_step2, cudaStream_t stream)
{
dim3 threads(32, 8, 1);
dim3 grid(1, 1, 1);
@ -503,16 +458,16 @@ namespace cv { namespace cuda { namespace device
switch(channels)
{
case 1: compute_data_cost<T, 1><<<grid, threads, 0, stream>>>(disp_selected_pyr, data_cost, h, w, level, nr_plane); break;
case 3: compute_data_cost<T, 3><<<grid, threads, 0, stream>>>(disp_selected_pyr, data_cost, h, w, level, nr_plane); break;
case 4: compute_data_cost<T, 4><<<grid, threads, 0, stream>>>(disp_selected_pyr, data_cost, h, w, level, nr_plane); break;
case 1: compute_data_cost<T, 1><<<grid, threads, 0, stream>>>(cleft, cright, cimg_step, disp_selected_pyr, data_cost, h, w, level, nr_plane, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2); break;
case 3: compute_data_cost<T, 3><<<grid, threads, 0, stream>>>(cleft, cright, cimg_step, disp_selected_pyr, data_cost, h, w, level, nr_plane, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2); break;
case 4: compute_data_cost<T, 4><<<grid, threads, 0, stream>>>(cleft, cright, cimg_step, disp_selected_pyr, data_cost, h, w, level, nr_plane, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2); break;
default: CV_Error(cv::Error::BadNumChannels, "Unsupported channels count");
}
}
template <typename T, int winsz>
void compute_data_cost_reduce_caller_(const T* disp_selected_pyr, T* data_cost, int rows, int cols,
int h, int w, int level, int nr_plane, int channels, cudaStream_t stream)
void compute_data_cost_reduce_caller_(const uchar *cleft, const uchar *cright, size_t cimg_step, const T* disp_selected_pyr, T* data_cost, int rows, int cols,
int h, int w, int level, int nr_plane, int channels, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step1, size_t disp_step2, cudaStream_t stream)
{
const int threadsNum = 256;
const size_t smem_size = threadsNum * sizeof(float);
@ -523,19 +478,20 @@ namespace cv { namespace cuda { namespace device
switch (channels)
{
case 1: compute_data_cost_reduce<T, winsz, 1><<<grid, threads, smem_size, stream>>>(disp_selected_pyr, data_cost, level, rows, cols, h, nr_plane); break;
case 3: compute_data_cost_reduce<T, winsz, 3><<<grid, threads, smem_size, stream>>>(disp_selected_pyr, data_cost, level, rows, cols, h, nr_plane); break;
case 4: compute_data_cost_reduce<T, winsz, 4><<<grid, threads, smem_size, stream>>>(disp_selected_pyr, data_cost, level, rows, cols, h, nr_plane); break;
case 1: compute_data_cost_reduce<T, winsz, 1><<<grid, threads, smem_size, stream>>>(cleft, cright, cimg_step, disp_selected_pyr, data_cost, level, rows, cols, h, nr_plane, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2); break;
case 3: compute_data_cost_reduce<T, winsz, 3><<<grid, threads, smem_size, stream>>>(cleft, cright, cimg_step, disp_selected_pyr, data_cost, level, rows, cols, h, nr_plane, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2); break;
case 4: compute_data_cost_reduce<T, winsz, 4><<<grid, threads, smem_size, stream>>>(cleft, cright, cimg_step, disp_selected_pyr, data_cost, level, rows, cols, h, nr_plane, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2); break;
default: CV_Error(cv::Error::BadNumChannels, "Unsupported channels count");
}
}
template<class T>
void compute_data_cost(const T* disp_selected_pyr, T* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, cudaStream_t stream)
void compute_data_cost(const uchar *cleft, const uchar *cright, size_t cimg_step, const T* disp_selected_pyr, T* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, float data_weight, float max_data_term,
int min_disp, cudaStream_t stream)
{
typedef void (*ComputeDataCostCaller)(const T* disp_selected_pyr, T* data_cost, int rows, int cols,
int h, int w, int level, int nr_plane, int channels, cudaStream_t stream);
typedef void (*ComputeDataCostCaller)(const uchar *cleft, const uchar *cright, size_t cimg_step, const T* disp_selected_pyr, T* data_cost, int rows, int cols,
int h, int w, int level, int nr_plane, int channels, float data_weight, float max_data_term, int min_disp, size_t msg_step, size_t disp_step1, size_t disp_step2, cudaStream_t stream);
static const ComputeDataCostCaller callers[] =
{
@ -546,22 +502,19 @@ namespace cv { namespace cuda { namespace device
size_t disp_step1 = msg_step * h;
size_t disp_step2 = msg_step * h2;
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step1, &disp_step1, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step2, &disp_step2, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cmsg_step, &msg_step, sizeof(size_t)) );
callers[level](disp_selected_pyr, data_cost, rows, cols, h, w, level, nr_plane, channels, stream);
callers[level](cleft, cright, cimg_step, disp_selected_pyr, data_cost, rows, cols, h, w, level, nr_plane, channels, data_weight, max_data_term, min_disp, msg_step, disp_step1, disp_step2, stream);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void compute_data_cost(const short* disp_selected_pyr, short* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, cudaStream_t stream);
template void compute_data_cost(const uchar *cleft, const uchar *cright, size_t cimg_step, const short* disp_selected_pyr, short* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, float data_weight, float max_data_term, int min_disp, cudaStream_t stream);
template void compute_data_cost(const float* disp_selected_pyr, float* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, cudaStream_t stream);
template void compute_data_cost(const uchar *cleft, const uchar *cright, size_t cimg_step, const float* disp_selected_pyr, float* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, float data_weight, float max_data_term, int min_disp, cudaStream_t stream);
///////////////////////////////////////////////////////////////
@ -574,7 +527,7 @@ namespace cv { namespace cuda { namespace device
const T* u_cur, const T* d_cur, const T* l_cur, const T* r_cur,
T* data_cost_selected, T* disparity_selected_new, T* data_cost_new,
const T* data_cost_cur, const T* disparity_selected_cur,
int nr_plane, int nr_plane2)
int nr_plane, int nr_plane2, size_t disp_step1, size_t disp_step2)
{
for(int i = 0; i < nr_plane; i++)
{
@ -582,7 +535,7 @@ namespace cv { namespace cuda { namespace device
int id = 0;
for(int j = 0; j < nr_plane2; j++)
{
T cur = data_cost_new[j * cdisp_step1];
T cur = data_cost_new[j * disp_step1];
if(cur < minimum)
{
minimum = cur;
@ -590,70 +543,72 @@ namespace cv { namespace cuda { namespace device
}
}
data_cost_selected[i * cdisp_step1] = data_cost_cur[id * cdisp_step1];
disparity_selected_new[i * cdisp_step1] = disparity_selected_cur[id * cdisp_step2];
data_cost_selected[i * disp_step1] = data_cost_cur[id * disp_step1];
disparity_selected_new[i * disp_step1] = disparity_selected_cur[id * disp_step2];
u_new[i * cdisp_step1] = u_cur[id * cdisp_step2];
d_new[i * cdisp_step1] = d_cur[id * cdisp_step2];
l_new[i * cdisp_step1] = l_cur[id * cdisp_step2];
r_new[i * cdisp_step1] = r_cur[id * cdisp_step2];
u_new[i * disp_step1] = u_cur[id * disp_step2];
d_new[i * disp_step1] = d_cur[id * disp_step2];
l_new[i * disp_step1] = l_cur[id * disp_step2];
r_new[i * disp_step1] = r_cur[id * disp_step2];
data_cost_new[id * cdisp_step1] = numeric_limits<T>::max();
data_cost_new[id * disp_step1] = numeric_limits<T>::max();
}
}
template <typename T>
__global__ void init_message(T* u_new_, T* d_new_, T* l_new_, T* r_new_,
__global__ void init_message(uchar *ctemp, T* u_new_, T* d_new_, T* l_new_, T* r_new_,
const T* u_cur_, const T* d_cur_, const T* l_cur_, const T* r_cur_,
T* selected_disp_pyr_new, const T* selected_disp_pyr_cur,
T* data_cost_selected_, const T* data_cost_,
int h, int w, int nr_plane, int h2, int w2, int nr_plane2)
int h, int w, int nr_plane, int h2, int w2, int nr_plane2,
size_t msg_step, size_t disp_step1, size_t disp_step2)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y < h && x < w)
{
const T* u_cur = u_cur_ + ::min(h2-1, y/2 + 1) * cmsg_step + x/2;
const T* d_cur = d_cur_ + ::max(0, y/2 - 1) * cmsg_step + x/2;
const T* l_cur = l_cur_ + (y/2) * cmsg_step + ::min(w2-1, x/2 + 1);
const T* r_cur = r_cur_ + (y/2) * cmsg_step + ::max(0, x/2 - 1);
const T* u_cur = u_cur_ + ::min(h2-1, y/2 + 1) * msg_step + x/2;
const T* d_cur = d_cur_ + ::max(0, y/2 - 1) * msg_step + x/2;
const T* l_cur = l_cur_ + (y/2) * msg_step + ::min(w2-1, x/2 + 1);
const T* r_cur = r_cur_ + (y/2) * msg_step + ::max(0, x/2 - 1);
T* data_cost_new = (T*)ctemp + y * cmsg_step + x;
T* data_cost_new = (T*)ctemp + y * msg_step + x;
const T* disparity_selected_cur = selected_disp_pyr_cur + y/2 * cmsg_step + x/2;
const T* data_cost = data_cost_ + y * cmsg_step + x;
const T* disparity_selected_cur = selected_disp_pyr_cur + y/2 * msg_step + x/2;
const T* data_cost = data_cost_ + y * msg_step + x;
for(int d = 0; d < nr_plane2; d++)
{
int idx2 = d * cdisp_step2;
int idx2 = d * disp_step2;
T val = data_cost[d * cdisp_step1] + u_cur[idx2] + d_cur[idx2] + l_cur[idx2] + r_cur[idx2];
data_cost_new[d * cdisp_step1] = val;
T val = data_cost[d * disp_step1] + u_cur[idx2] + d_cur[idx2] + l_cur[idx2] + r_cur[idx2];
data_cost_new[d * disp_step1] = val;
}
T* data_cost_selected = data_cost_selected_ + y * cmsg_step + x;
T* disparity_selected_new = selected_disp_pyr_new + y * cmsg_step + x;
T* data_cost_selected = data_cost_selected_ + y * msg_step + x;
T* disparity_selected_new = selected_disp_pyr_new + y * msg_step + x;
T* u_new = u_new_ + y * cmsg_step + x;
T* d_new = d_new_ + y * cmsg_step + x;
T* l_new = l_new_ + y * cmsg_step + x;
T* r_new = r_new_ + y * cmsg_step + x;
T* u_new = u_new_ + y * msg_step + x;
T* d_new = d_new_ + y * msg_step + x;
T* l_new = l_new_ + y * msg_step + x;
T* r_new = r_new_ + y * msg_step + x;
u_cur = u_cur_ + y/2 * cmsg_step + x/2;
d_cur = d_cur_ + y/2 * cmsg_step + x/2;
l_cur = l_cur_ + y/2 * cmsg_step + x/2;
r_cur = r_cur_ + y/2 * cmsg_step + x/2;
u_cur = u_cur_ + y/2 * msg_step + x/2;
d_cur = d_cur_ + y/2 * msg_step + x/2;
l_cur = l_cur_ + y/2 * msg_step + x/2;
r_cur = r_cur_ + y/2 * msg_step + x/2;
get_first_k_element_increase(u_new, d_new, l_new, r_new, u_cur, d_cur, l_cur, r_cur,
data_cost_selected, disparity_selected_new, data_cost_new,
data_cost, disparity_selected_cur, nr_plane, nr_plane2);
data_cost, disparity_selected_cur, nr_plane, nr_plane2,
disp_step1, disp_step2);
}
}
template<class T>
void init_message(T* u_new, T* d_new, T* l_new, T* r_new,
void init_message(uchar *ctemp, T* u_new, T* d_new, T* l_new, T* r_new,
const T* u_cur, const T* d_cur, const T* l_cur, const T* r_cur,
T* selected_disp_pyr_new, const T* selected_disp_pyr_cur,
T* data_cost_selected, const T* data_cost, size_t msg_step,
@ -662,9 +617,6 @@ namespace cv { namespace cuda { namespace device
size_t disp_step1 = msg_step * h;
size_t disp_step2 = msg_step * h2;
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step1, &disp_step1, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step2, &disp_step2, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cmsg_step, &msg_step, sizeof(size_t)) );
dim3 threads(32, 8, 1);
dim3 grid(1, 1, 1);
@ -672,11 +624,12 @@ namespace cv { namespace cuda { namespace device
grid.x = divUp(w, threads.x);
grid.y = divUp(h, threads.y);
init_message<<<grid, threads, 0, stream>>>(u_new, d_new, l_new, r_new,
init_message<<<grid, threads, 0, stream>>>(ctemp, u_new, d_new, l_new, r_new,
u_cur, d_cur, l_cur, r_cur,
selected_disp_pyr_new, selected_disp_pyr_cur,
data_cost_selected, data_cost,
h, w, nr_plane, h2, w2, nr_plane2);
h, w, nr_plane, h2, w2, nr_plane2,
msg_step, disp_step1, disp_step2);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
@ -684,13 +637,13 @@ namespace cv { namespace cuda { namespace device
}
template void init_message(short* u_new, short* d_new, short* l_new, short* r_new,
template void init_message(uchar *ctemp, short* u_new, short* d_new, short* l_new, short* r_new,
const short* u_cur, const short* d_cur, const short* l_cur, const short* r_cur,
short* selected_disp_pyr_new, const short* selected_disp_pyr_cur,
short* data_cost_selected, const short* data_cost, size_t msg_step,
int h, int w, int nr_plane, int h2, int w2, int nr_plane2, cudaStream_t stream);
template void init_message(float* u_new, float* d_new, float* l_new, float* r_new,
template void init_message(uchar *ctemp, float* u_new, float* d_new, float* l_new, float* r_new,
const float* u_cur, const float* d_cur, const float* l_cur, const float* r_cur,
float* selected_disp_pyr_new, const float* selected_disp_pyr_cur,
float* data_cost_selected, const float* data_cost, size_t msg_step,
@ -702,13 +655,14 @@ namespace cv { namespace cuda { namespace device
template <typename T>
__device__ void message_per_pixel(const T* data, T* msg_dst, const T* msg1, const T* msg2, const T* msg3,
const T* dst_disp, const T* src_disp, int nr_plane, volatile T* temp)
const T* dst_disp, const T* src_disp, int nr_plane, int max_disc_term, float disc_single_jump, volatile T* temp,
size_t disp_step)
{
T minimum = numeric_limits<T>::max();
for(int d = 0; d < nr_plane; d++)
{
int idx = d * cdisp_step1;
int idx = d * disp_step;
T val = data[idx] + msg1[idx] + msg2[idx] + msg3[idx];
if(val < minimum)
@ -720,55 +674,53 @@ namespace cv { namespace cuda { namespace device
float sum = 0;
for(int d = 0; d < nr_plane; d++)
{
float cost_min = minimum + cmax_disc_term;
T src_disp_reg = src_disp[d * cdisp_step1];
float cost_min = minimum + max_disc_term;
T src_disp_reg = src_disp[d * disp_step];
for(int d2 = 0; d2 < nr_plane; d2++)
cost_min = fmin(cost_min, msg_dst[d2 * cdisp_step1] + cdisc_single_jump * ::abs(dst_disp[d2 * cdisp_step1] - src_disp_reg));
cost_min = fmin(cost_min, msg_dst[d2 * disp_step] + disc_single_jump * ::abs(dst_disp[d2 * disp_step] - src_disp_reg));
temp[d * cdisp_step1] = saturate_cast<T>(cost_min);
temp[d * disp_step] = saturate_cast<T>(cost_min);
sum += cost_min;
}
sum /= nr_plane;
for(int d = 0; d < nr_plane; d++)
msg_dst[d * cdisp_step1] = saturate_cast<T>(temp[d * cdisp_step1] - sum);
msg_dst[d * disp_step] = saturate_cast<T>(temp[d * disp_step] - sum);
}
template <typename T>
__global__ void compute_message(T* u_, T* d_, T* l_, T* r_, const T* data_cost_selected, const T* selected_disp_pyr_cur, int h, int w, int nr_plane, int i)
__global__ void compute_message(uchar *ctemp, T* u_, T* d_, T* l_, T* r_, const T* data_cost_selected, const T* selected_disp_pyr_cur, int h, int w, int nr_plane, int i, int max_disc_term, float disc_single_jump, size_t msg_step, size_t disp_step)
{
int y = blockIdx.y * blockDim.y + threadIdx.y;
int x = ((blockIdx.x * blockDim.x + threadIdx.x) << 1) + ((y + i) & 1);
if (y > 0 && y < h - 1 && x > 0 && x < w - 1)
{
const T* data = data_cost_selected + y * cmsg_step + x;
const T* data = data_cost_selected + y * msg_step + x;
T* u = u_ + y * cmsg_step + x;
T* d = d_ + y * cmsg_step + x;
T* l = l_ + y * cmsg_step + x;
T* r = r_ + y * cmsg_step + x;
T* u = u_ + y * msg_step + x;
T* d = d_ + y * msg_step + x;
T* l = l_ + y * msg_step + x;
T* r = r_ + y * msg_step + x;
const T* disp = selected_disp_pyr_cur + y * cmsg_step + x;
const T* disp = selected_disp_pyr_cur + y * msg_step + x;
T* temp = (T*)ctemp + y * cmsg_step + x;
T* temp = (T*)ctemp + y * msg_step + x;
message_per_pixel(data, u, r - 1, u + cmsg_step, l + 1, disp, disp - cmsg_step, nr_plane, temp);
message_per_pixel(data, d, d - cmsg_step, r - 1, l + 1, disp, disp + cmsg_step, nr_plane, temp);
message_per_pixel(data, l, u + cmsg_step, d - cmsg_step, l + 1, disp, disp - 1, nr_plane, temp);
message_per_pixel(data, r, u + cmsg_step, d - cmsg_step, r - 1, disp, disp + 1, nr_plane, temp);
message_per_pixel(data, u, r - 1, u + msg_step, l + 1, disp, disp - msg_step, nr_plane, max_disc_term, disc_single_jump, temp, disp_step);
message_per_pixel(data, d, d - msg_step, r - 1, l + 1, disp, disp + msg_step, nr_plane, max_disc_term, disc_single_jump, temp, disp_step);
message_per_pixel(data, l, u + msg_step, d - msg_step, l + 1, disp, disp - 1, nr_plane, max_disc_term, disc_single_jump, temp, disp_step);
message_per_pixel(data, r, u + msg_step, d - msg_step, r - 1, disp, disp + 1, nr_plane, max_disc_term, disc_single_jump, temp, disp_step);
}
}
template<class T>
void calc_all_iterations(T* u, T* d, T* l, T* r, const T* data_cost_selected,
const T* selected_disp_pyr_cur, size_t msg_step, int h, int w, int nr_plane, int iters, cudaStream_t stream)
void calc_all_iterations(uchar *ctemp, T* u, T* d, T* l, T* r, const T* data_cost_selected,
const T* selected_disp_pyr_cur, size_t msg_step, int h, int w, int nr_plane, int iters, int max_disc_term, float disc_single_jump, cudaStream_t stream)
{
size_t disp_step = msg_step * h;
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step1, &disp_step, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cmsg_step, &msg_step, sizeof(size_t)) );
dim3 threads(32, 8, 1);
dim3 grid(1, 1, 1);
@ -778,18 +730,18 @@ namespace cv { namespace cuda { namespace device
for(int t = 0; t < iters; ++t)
{
compute_message<<<grid, threads, 0, stream>>>(u, d, l, r, data_cost_selected, selected_disp_pyr_cur, h, w, nr_plane, t & 1);
compute_message<<<grid, threads, 0, stream>>>(ctemp, u, d, l, r, data_cost_selected, selected_disp_pyr_cur, h, w, nr_plane, t & 1, max_disc_term, disc_single_jump, msg_step, disp_step);
cudaSafeCall( cudaGetLastError() );
}
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
};
template void calc_all_iterations(short* u, short* d, short* l, short* r, const short* data_cost_selected, const short* selected_disp_pyr_cur, size_t msg_step,
int h, int w, int nr_plane, int iters, cudaStream_t stream);
template void calc_all_iterations(uchar *ctemp, short* u, short* d, short* l, short* r, const short* data_cost_selected, const short* selected_disp_pyr_cur, size_t msg_step,
int h, int w, int nr_plane, int iters, int max_disc_term, float disc_single_jump, cudaStream_t stream);
template void calc_all_iterations(float* u, float* d, float* l, float* r, const float* data_cost_selected, const float* selected_disp_pyr_cur, size_t msg_step,
int h, int w, int nr_plane, int iters, cudaStream_t stream);
template void calc_all_iterations(uchar *ctemp, float* u, float* d, float* l, float* r, const float* data_cost_selected, const float* selected_disp_pyr_cur, size_t msg_step,
int h, int w, int nr_plane, int iters, int max_disc_term, float disc_single_jump, cudaStream_t stream);
///////////////////////////////////////////////////////////////
@ -800,26 +752,26 @@ namespace cv { namespace cuda { namespace device
template <typename T>
__global__ void compute_disp(const T* u_, const T* d_, const T* l_, const T* r_,
const T* data_cost_selected, const T* disp_selected_pyr,
PtrStepSz<short> disp, int nr_plane)
PtrStepSz<short> disp, int nr_plane, size_t msg_step, size_t disp_step)
{
int x = blockIdx.x * blockDim.x + threadIdx.x;
int y = blockIdx.y * blockDim.y + threadIdx.y;
if (y > 0 && y < disp.rows - 1 && x > 0 && x < disp.cols - 1)
{
const T* data = data_cost_selected + y * cmsg_step + x;
const T* disp_selected = disp_selected_pyr + y * cmsg_step + x;
const T* data = data_cost_selected + y * msg_step + x;
const T* disp_selected = disp_selected_pyr + y * msg_step + x;
const T* u = u_ + (y+1) * cmsg_step + (x+0);
const T* d = d_ + (y-1) * cmsg_step + (x+0);
const T* l = l_ + (y+0) * cmsg_step + (x+1);
const T* r = r_ + (y+0) * cmsg_step + (x-1);
const T* u = u_ + (y+1) * msg_step + (x+0);
const T* d = d_ + (y-1) * msg_step + (x+0);
const T* l = l_ + (y+0) * msg_step + (x+1);
const T* r = r_ + (y+0) * msg_step + (x-1);
int best = 0;
T best_val = numeric_limits<T>::max();
for (int i = 0; i < nr_plane; ++i)
{
int idx = i * cdisp_step1;
int idx = i * disp_step;
T val = data[idx]+ u[idx] + d[idx] + l[idx] + r[idx];
if (val < best_val)
@ -837,8 +789,6 @@ namespace cv { namespace cuda { namespace device
const PtrStepSz<short>& disp, int nr_plane, cudaStream_t stream)
{
size_t disp_step = disp.rows * msg_step;
cudaSafeCall( cudaMemcpyToSymbol(cdisp_step1, &disp_step, sizeof(size_t)) );
cudaSafeCall( cudaMemcpyToSymbol(cmsg_step, &msg_step, sizeof(size_t)) );
dim3 threads(32, 8, 1);
dim3 grid(1, 1, 1);
@ -846,7 +796,7 @@ namespace cv { namespace cuda { namespace device
grid.x = divUp(disp.cols, threads.x);
grid.y = divUp(disp.rows, threads.y);
compute_disp<<<grid, threads, 0, stream>>>(u, d, l, r, data_cost_selected, disp_selected, disp, nr_plane);
compute_disp<<<grid, threads, 0, stream>>>(u, d, l, r, data_cost_selected, disp_selected, disp, nr_plane, msg_step, disp_step);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)

29
modules/cudastereo/src/cuda/stereocsbp.hpp Normal file
View File

@ -0,0 +1,29 @@
namespace cv { namespace cuda { namespace device
{
namespace stereocsbp
{
template<class T>
void init_data_cost(const uchar *left, const uchar *right, uchar *ctemp, size_t cimg_step, int rows, int cols, T* disp_selected_pyr, T* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, float data_weight, float max_data_term, int min_disp, bool use_local_init_data_cost, cudaStream_t stream);
template<class T>
void compute_data_cost(const uchar *left, const uchar *right, size_t cimg_step, const T* disp_selected_pyr, T* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, float data_weight, float max_data_term,
int min_disp, cudaStream_t stream);
template<class T>
void init_message(uchar *ctemp, T* u_new, T* d_new, T* l_new, T* r_new,
const T* u_cur, const T* d_cur, const T* l_cur, const T* r_cur,
T* selected_disp_pyr_new, const T* selected_disp_pyr_cur,
T* data_cost_selected, const T* data_cost, size_t msg_step,
int h, int w, int nr_plane, int h2, int w2, int nr_plane2, cudaStream_t stream);
template<class T>
void calc_all_iterations(uchar *ctemp, T* u, T* d, T* l, T* r, const T* data_cost_selected,
const T* selected_disp_pyr_cur, size_t msg_step, int h, int w, int nr_plane, int iters, int max_disc_term, float disc_single_jump, cudaStream_t stream);
template<class T>
void compute_disp(const T* u, const T* d, const T* l, const T* r, const T* data_cost_selected, const T* disp_selected, size_t msg_step,
const PtrStepSz<short>& disp, int nr_plane, cudaStream_t stream);
}
}}}

15
modules/cudastereo/src/disparity_bilateral_filter.cpp
View File

@ -51,16 +51,7 @@ Ptr<cuda::DisparityBilateralFilter> cv::cuda::createDisparityBilateralFilter(int
#else /* !defined (HAVE_CUDA) */
namespace cv { namespace cuda { namespace device
{
namespace disp_bilateral_filter
{
void disp_load_constants(float* table_color, PtrStepSzf table_space, int ndisp, int radius, short edge_disc, short max_disc);
template<typename T>
void disp_bilateral_filter(PtrStepSz<T> disp, PtrStepSzb img, int channels, int iters, cudaStream_t stream);
}
}}}
#include "cuda/disparity_bilateral_filter.hpp"
namespace
{
@ -165,7 +156,7 @@ namespace
const short edge_disc = std::max<short>(short(1), short(ndisp * edge_threshold + 0.5));
const short max_disc = short(ndisp * max_disc_threshold + 0.5);
disp_load_constants(table_color.ptr<float>(), table_space, ndisp, radius, edge_disc, max_disc);
size_t table_space_step = table_space.step / sizeof(float);
_dst.create(disp.size(), disp.type());
GpuMat dst = _dst.getGpuMat();
@ -173,7 +164,7 @@ namespace
if (dst.data != disp.data)
disp.copyTo(dst, stream);
disp_bilateral_filter<T>(dst, img, img.channels(), iters, StreamAccessor::getStream(stream));
disp_bilateral_filter<T>(dst, img, img.channels(), iters, table_color.ptr<float>(), (float *)table_space.data, table_space_step, radius, edge_disc, max_disc, StreamAccessor::getStream(stream));
}
void DispBilateralFilterImpl::apply(InputArray _disp, InputArray _image, OutputArray dst, Stream& stream)

64
modules/cudastereo/src/stereocsbp.cpp
View File

@ -53,37 +53,7 @@ Ptr<cuda::StereoConstantSpaceBP> cv::cuda::createStereoConstantSpaceBP(int, int,
#else /* !defined (HAVE_CUDA) */
namespace cv { namespace cuda { namespace device
{
namespace stereocsbp
{
void load_constants(int ndisp, float max_data_term, float data_weight, float max_disc_term, float disc_single_jump, int min_disp_th,
const PtrStepSzb& left, const PtrStepSzb& right, const PtrStepSzb& temp);
template<class T>
void init_data_cost(int rows, int cols, T* disp_selected_pyr, T* data_cost_selected, size_t msg_step,
int h, int w, int level, int nr_plane, int ndisp, int channels, bool use_local_init_data_cost, cudaStream_t stream);
template<class T>
void compute_data_cost(const T* disp_selected_pyr, T* data_cost, size_t msg_step,
int rows, int cols, int h, int w, int h2, int level, int nr_plane, int channels, cudaStream_t stream);
template<class T>
void init_message(T* u_new, T* d_new, T* l_new, T* r_new,
const T* u_cur, const T* d_cur, const T* l_cur, const T* r_cur,
T* selected_disp_pyr_new, const T* selected_disp_pyr_cur,
T* data_cost_selected, const T* data_cost, size_t msg_step,
int h, int w, int nr_plane, int h2, int w2, int nr_plane2, cudaStream_t stream);
template<class T>
void calc_all_iterations(T* u, T* d, T* l, T* r, const T* data_cost_selected,
const T* selected_disp_pyr_cur, size_t msg_step, int h, int w, int nr_plane, int iters, cudaStream_t stream);
template<class T>
void compute_disp(const T* u, const T* d, const T* l, const T* r, const T* data_cost_selected, const T* disp_selected, size_t msg_step,
const PtrStepSz<short>& disp, int nr_plane, cudaStream_t stream);
}
}}}
#include "cuda/stereocsbp.hpp"
namespace
{
@ -252,8 +222,6 @@ namespace
////////////////////////////////////////////////////////////////////////////
// Compute
load_constants(ndisp_, max_data_term_, data_weight_, max_disc_term_, disc_single_jump_, min_disp_th_, left, right, temp_);
l[0].setTo(0, _stream);
d[0].setTo(0, _stream);
r[0].setTo(0, _stream);
@ -275,17 +243,18 @@ namespace
{
if (i == levels_ - 1)
{
init_data_cost(left.rows, left.cols, disp_selected_pyr[cur_idx].ptr<float>(), data_cost_selected.ptr<float>(),
elem_step, rows_pyr[i], cols_pyr[i], i, nr_plane_pyr[i], ndisp_, left.channels(), use_local_init_data_cost_, stream);
init_data_cost(left.ptr<uchar>(), right.ptr<uchar>(), temp_.ptr<uchar>(), left.step, left.rows, left.cols, disp_selected_pyr[cur_idx].ptr<float>(), data_cost_selected.ptr<float>(),
elem_step, rows_pyr[i], cols_pyr[i], i, nr_plane_pyr[i], ndisp_, left.channels(), data_weight_, max_data_term_, min_disp_th_, use_local_init_data_cost_, stream);
}
else
{
compute_data_cost(disp_selected_pyr[cur_idx].ptr<float>(), data_cost.ptr<float>(), elem_step,
left.rows, left.cols, rows_pyr[i], cols_pyr[i], rows_pyr[i+1], i, nr_plane_pyr[i+1], left.channels(), stream);
compute_data_cost(left.ptr<uchar>(), right.ptr<uchar>(), left.step, disp_selected_pyr[cur_idx].ptr<float>(), data_cost.ptr<float>(), elem_step,
left.rows, left.cols, rows_pyr[i], cols_pyr[i], rows_pyr[i+1], i, nr_plane_pyr[i+1], left.channels(), data_weight_, max_data_term_, min_disp_th_, stream);
int new_idx = (cur_idx + 1) & 1;
init_message(u[new_idx].ptr<float>(), d[new_idx].ptr<float>(), l[new_idx].ptr<float>(), r[new_idx].ptr<float>(),
init_message(temp_.ptr<uchar>(),
u[new_idx].ptr<float>(), d[new_idx].ptr<float>(), l[new_idx].ptr<float>(), r[new_idx].ptr<float>(),
u[cur_idx].ptr<float>(), d[cur_idx].ptr<float>(), l[cur_idx].ptr<float>(), r[cur_idx].ptr<float>(),
disp_selected_pyr[new_idx].ptr<float>(), disp_selected_pyr[cur_idx].ptr<float>(),
data_cost_selected.ptr<float>(), data_cost.ptr<float>(), elem_step, rows_pyr[i],
@ -294,9 +263,9 @@ namespace
cur_idx = new_idx;
}
calc_all_iterations(u[cur_idx].ptr<float>(), d[cur_idx].ptr<float>(), l[cur_idx].ptr<float>(), r[cur_idx].ptr<float>(),
calc_all_iterations(temp_.ptr<uchar>(), u[cur_idx].ptr<float>(), d[cur_idx].ptr<float>(), l[cur_idx].ptr<float>(), r[cur_idx].ptr<float>(),
data_cost_selected.ptr<float>(), disp_selected_pyr[cur_idx].ptr<float>(), elem_step,
rows_pyr[i], cols_pyr[i], nr_plane_pyr[i], iters_, stream);
rows_pyr[i], cols_pyr[i], nr_plane_pyr[i], iters_, max_disc_term_, disc_single_jump_, stream);
}
}
else
@ -305,17 +274,18 @@ namespace
{
if (i == levels_ - 1)
{
init_data_cost(left.rows, left.cols, disp_selected_pyr[cur_idx].ptr<short>(), data_cost_selected.ptr<short>(),
elem_step, rows_pyr[i], cols_pyr[i], i, nr_plane_pyr[i], ndisp_, left.channels(), use_local_init_data_cost_, stream);
init_data_cost(left.ptr<uchar>(), right.ptr<uchar>(), temp_.ptr<uchar>(), left.step, left.rows, left.cols, disp_selected_pyr[cur_idx].ptr<short>(), data_cost_selected.ptr<short>(),
elem_step, rows_pyr[i], cols_pyr[i], i, nr_plane_pyr[i], ndisp_, left.channels(), data_weight_, max_data_term_, min_disp_th_, use_local_init_data_cost_, stream);
}
else
{
compute_data_cost(disp_selected_pyr[cur_idx].ptr<short>(), data_cost.ptr<short>(), elem_step,
left.rows, left.cols, rows_pyr[i], cols_pyr[i], rows_pyr[i+1], i, nr_plane_pyr[i+1], left.channels(), stream);
compute_data_cost(left.ptr<uchar>(), right.ptr<uchar>(), left.step, disp_selected_pyr[cur_idx].ptr<short>(), data_cost.ptr<short>(), elem_step,
left.rows, left.cols, rows_pyr[i], cols_pyr[i], rows_pyr[i+1], i, nr_plane_pyr[i+1], left.channels(), data_weight_, max_data_term_, min_disp_th_, stream);
int new_idx = (cur_idx + 1) & 1;
init_message(u[new_idx].ptr<short>(), d[new_idx].ptr<short>(), l[new_idx].ptr<short>(), r[new_idx].ptr<short>(),
init_message(temp_.ptr<uchar>(),
u[new_idx].ptr<short>(), d[new_idx].ptr<short>(), l[new_idx].ptr<short>(), r[new_idx].ptr<short>(),
u[cur_idx].ptr<short>(), d[cur_idx].ptr<short>(), l[cur_idx].ptr<short>(), r[cur_idx].ptr<short>(),
disp_selected_pyr[new_idx].ptr<short>(), disp_selected_pyr[cur_idx].ptr<short>(),
data_cost_selected.ptr<short>(), data_cost.ptr<short>(), elem_step, rows_pyr[i],
@ -324,9 +294,9 @@ namespace
cur_idx = new_idx;
}
calc_all_iterations(u[cur_idx].ptr<short>(), d[cur_idx].ptr<short>(), l[cur_idx].ptr<short>(), r[cur_idx].ptr<short>(),
calc_all_iterations(temp_.ptr<uchar>(), u[cur_idx].ptr<short>(), d[cur_idx].ptr<short>(), l[cur_idx].ptr<short>(), r[cur_idx].ptr<short>(),
data_cost_selected.ptr<short>(), disp_selected_pyr[cur_idx].ptr<short>(), elem_step,
rows_pyr[i], cols_pyr[i], nr_plane_pyr[i], iters_, stream);
rows_pyr[i], cols_pyr[i], nr_plane_pyr[i], iters_, max_disc_term_, disc_single_jump_, stream);
}
}

44
modules/features2d/doc/feature_detection_and_description.rst
View File

@ -31,7 +31,7 @@ Detects corners using the FAST algorithm
Detects corners using the FAST algorithm by [Rosten06]_.
..note:: In Python API, types are given as ``cv2.FAST_FEATURE_DETECTOR_TYPE_5_8``, ``cv2.FAST_FEATURE_DETECTOR_TYPE_7_12`` and ``cv2.FAST_FEATURE_DETECTOR_TYPE_9_16``. For corner detection, use ``cv2.FAST.detect()`` method.
.. note:: In Python API, types are given as ``cv2.FAST_FEATURE_DETECTOR_TYPE_5_8``, ``cv2.FAST_FEATURE_DETECTOR_TYPE_7_12`` and ``cv2.FAST_FEATURE_DETECTOR_TYPE_9_16``. For corner detection, use ``cv2.FAST.detect()`` method.
.. [Rosten06] E. Rosten. Machine Learning for High-speed Corner Detection, 2006.
@ -254,7 +254,17 @@ KAZE
----
.. ocv:class:: KAZE : public Feature2D
Class implementing the KAZE keypoint detector and descriptor extractor, described in [ABD12]_.
Class implementing the KAZE keypoint detector and descriptor extractor, described in [ABD12]_. ::
class CV_EXPORTS_W KAZE : public Feature2D
{
public:
CV_WRAP KAZE();
CV_WRAP explicit KAZE(bool extended, bool upright, float threshold = 0.001f,
int octaves = 4, int sublevels = 4, int diffusivity = DIFF_PM_G2);
};
.. note:: AKAZE descriptor can only be used with KAZE or AKAZE keypoints
.. [ABD12] KAZE Features. Pablo F. Alcantarilla, Adrien Bartoli and Andrew J. Davison. In European Conference on Computer Vision (ECCV), Fiorenze, Italy, October 2012.
@ -262,12 +272,14 @@ KAZE::KAZE
----------
The KAZE constructor
.. ocv:function:: KAZE::KAZE(bool extended, bool upright)
.. ocv:function:: KAZE::KAZE(bool extended, bool upright, float threshold, int octaves, int sublevels, int diffusivity)
:param extended: Set to enable extraction of extended (128-byte) descriptor.
:param upright: Set to enable use of upright descriptors (non rotation-invariant).
:param threshold: Detector response threshold to accept point
:param octaves: Maximum octave evolution of the image
:param sublevels: Default number of sublevels per scale level
:param diffusivity: Diffusivity type. DIFF_PM_G1, DIFF_PM_G2, DIFF_WEICKERT or DIFF_CHARBONNIER
AKAZE
-----
@ -278,25 +290,25 @@ Class implementing the AKAZE keypoint detector and descriptor extractor, describ
class CV_EXPORTS_W AKAZE : public Feature2D
{
public:
/// AKAZE Descriptor Type
enum DESCRIPTOR_TYPE {
DESCRIPTOR_KAZE_UPRIGHT = 2, ///< Upright descriptors, not invariant to rotation
DESCRIPTOR_KAZE = 3,
DESCRIPTOR_MLDB_UPRIGHT = 4, ///< Upright descriptors, not invariant to rotation
DESCRIPTOR_MLDB = 5
};
CV_WRAP AKAZE();
explicit AKAZE(DESCRIPTOR_TYPE descriptor_type, int descriptor_size = 0, int descriptor_channels = 3);
CV_WRAP explicit AKAZE(int descriptor_type, int descriptor_size = 0, int descriptor_channels = 3,
float threshold = 0.001f, int octaves = 4, int sublevels = 4, int diffusivity = DIFF_PM_G2);
};
.. note:: AKAZE descriptor can only be used with KAZE or AKAZE keypoints
.. [ANB13] Fast Explicit Diffusion for Accelerated Features in Nonlinear Scale Spaces. Pablo F. Alcantarilla, Jesús Nuevo and Adrien Bartoli. In British Machine Vision Conference (BMVC), Bristol, UK, September 2013.
AKAZE::AKAZE
------------
The AKAZE constructor
.. ocv:function:: AKAZE::AKAZE(DESCRIPTOR_TYPE descriptor_type, int descriptor_size = 0, int descriptor_channels = 3)
.. ocv:function:: AKAZE::AKAZE(int descriptor_type, int descriptor_size, int descriptor_channels, float threshold, int octaves, int sublevels, int diffusivity)
:param descriptor_type: Type of the extracted descriptor.
:param descriptor_type: Type of the extracted descriptor: DESCRIPTOR_KAZE, DESCRIPTOR_KAZE_UPRIGHT, DESCRIPTOR_MLDB or DESCRIPTOR_MLDB_UPRIGHT.
:param descriptor_size: Size of the descriptor in bits. 0 -> Full size
:param descriptor_channels: Number of channels in the descriptor (1, 2, 3).
:param descriptor_channels: Number of channels in the descriptor (1, 2, 3)
:param threshold: Detector response threshold to accept point
:param octaves: Maximum octave evolution of the image
:param sublevels: Default number of sublevels per scale level
:param diffusivity: Diffusivity type. DIFF_PM_G1, DIFF_PM_G2, DIFF_WEICKERT or DIFF_CHARBONNIER

39
modules/features2d/include/opencv2/features2d.hpp
View File

@ -895,6 +895,22 @@ protected:
PixelTestFn test_fn_;
};
// KAZE/AKAZE diffusivity
enum {
DIFF_PM_G1 = 0,
DIFF_PM_G2 = 1,
DIFF_WEICKERT = 2,
DIFF_CHARBONNIER = 3
};
// AKAZE descriptor type
enum {
DESCRIPTOR_KAZE_UPRIGHT = 2, ///< Upright descriptors, not invariant to rotation
DESCRIPTOR_KAZE = 3,
DESCRIPTOR_MLDB_UPRIGHT = 4, ///< Upright descriptors, not invariant to rotation
DESCRIPTOR_MLDB = 5
};
/*!
KAZE implementation
*/
@ -902,7 +918,8 @@ class CV_EXPORTS_W KAZE : public Feature2D
{
public:
CV_WRAP KAZE();
CV_WRAP explicit KAZE(bool extended, bool upright);
CV_WRAP explicit KAZE(bool extended, bool upright, float threshold = 0.001f,
int octaves = 4, int sublevels = 4, int diffusivity = DIFF_PM_G2);
virtual ~KAZE();
@ -928,6 +945,10 @@ protected:
CV_PROP bool extended;
CV_PROP bool upright;
CV_PROP float threshold;
CV_PROP int octaves;
CV_PROP int sublevels;
CV_PROP int diffusivity;
};
/*!
@ -936,16 +957,9 @@ AKAZE implementation
class CV_EXPORTS_W AKAZE : public Feature2D
{
public:
/// AKAZE Descriptor Type
enum DESCRIPTOR_TYPE {
DESCRIPTOR_KAZE_UPRIGHT = 2, ///< Upright descriptors, not invariant to rotation
DESCRIPTOR_KAZE = 3,
DESCRIPTOR_MLDB_UPRIGHT = 4, ///< Upright descriptors, not invariant to rotation
DESCRIPTOR_MLDB = 5
};
CV_WRAP AKAZE();
explicit AKAZE(DESCRIPTOR_TYPE descriptor_type, int descriptor_size = 0, int descriptor_channels = 3);
CV_WRAP explicit AKAZE(int descriptor_type, int descriptor_size = 0, int descriptor_channels = 3,
float threshold = 0.001f, int octaves = 4, int sublevels = 4, int diffusivity = DIFF_PM_G2);
virtual ~AKAZE();
@ -973,7 +987,10 @@ protected:
CV_PROP int descriptor;
CV_PROP int descriptor_channels;
CV_PROP int descriptor_size;
CV_PROP float threshold;
CV_PROP int octaves;
CV_PROP int sublevels;
CV_PROP int diffusivity;
};
/****************************************************************************************\
* Distance *

52
modules/features2d/src/akaze.cpp
View File

@ -49,7 +49,10 @@ http://www.robesafe.com/personal/pablo.alcantarilla/papers/Alcantarilla13bmvc.pd
*/
#include "precomp.hpp"
#include "akaze/AKAZEFeatures.h"
#include "kaze/AKAZEFeatures.h"
#include <iostream>
using namespace std;
namespace cv
{
@ -57,13 +60,22 @@ namespace cv
: descriptor(DESCRIPTOR_MLDB)
, descriptor_channels(3)
, descriptor_size(0)
, threshold(0.001f)
, octaves(4)
, sublevels(4)
, diffusivity(DIFF_PM_G2)
{
}
AKAZE::AKAZE(DESCRIPTOR_TYPE _descriptor_type, int _descriptor_size, int _descriptor_channels)
AKAZE::AKAZE(int _descriptor_type, int _descriptor_size, int _descriptor_channels,
float _threshold, int _octaves, int _sublevels, int _diffusivity)
: descriptor(_descriptor_type)
, descriptor_channels(_descriptor_channels)
, descriptor_size(_descriptor_size)
, threshold(_threshold)
, octaves(_octaves)
, sublevels(_sublevels)
, diffusivity(_diffusivity)
{
}
@ -78,12 +90,12 @@ namespace cv
{
switch (descriptor)
{
case cv::AKAZE::DESCRIPTOR_KAZE:
case cv::AKAZE::DESCRIPTOR_KAZE_UPRIGHT:
case cv::DESCRIPTOR_KAZE:
case cv::DESCRIPTOR_KAZE_UPRIGHT:
return 64;
case cv::AKAZE::DESCRIPTOR_MLDB:
case cv::AKAZE::DESCRIPTOR_MLDB_UPRIGHT:
case cv::DESCRIPTOR_MLDB:
case cv::DESCRIPTOR_MLDB_UPRIGHT:
// We use the full length binary descriptor -> 486 bits
if (descriptor_size == 0)
{
@ -106,12 +118,12 @@ namespace cv
{
switch (descriptor)
{
case cv::AKAZE::DESCRIPTOR_KAZE:
case cv::AKAZE::DESCRIPTOR_KAZE_UPRIGHT:
case cv::DESCRIPTOR_KAZE:
case cv::DESCRIPTOR_KAZE_UPRIGHT:
return CV_32F;
case cv::AKAZE::DESCRIPTOR_MLDB:
case cv::AKAZE::DESCRIPTOR_MLDB_UPRIGHT:
case cv::DESCRIPTOR_MLDB:
case cv::DESCRIPTOR_MLDB_UPRIGHT:
return CV_8U;
default:
@ -124,12 +136,12 @@ namespace cv
{
switch (descriptor)
{
case cv::AKAZE::DESCRIPTOR_KAZE:
case cv::AKAZE::DESCRIPTOR_KAZE_UPRIGHT:
case cv::DESCRIPTOR_KAZE:
case cv::DESCRIPTOR_KAZE_UPRIGHT:
return cv::NORM_L2;
case cv::AKAZE::DESCRIPTOR_MLDB:
case cv::AKAZE::DESCRIPTOR_MLDB_UPRIGHT:
case cv::DESCRIPTOR_MLDB:
case cv::DESCRIPTOR_MLDB_UPRIGHT:
return cv::NORM_HAMMING;
default:
@ -153,11 +165,15 @@ namespace cv
cv::Mat& desc = descriptors.getMatRef();
AKAZEOptions options;
options.descriptor = static_cast<DESCRIPTOR_TYPE>(descriptor);
options.descriptor = descriptor;
options.descriptor_channels = descriptor_channels;
options.descriptor_size = descriptor_size;
options.img_width = img.cols;
options.img_height = img.rows;
options.dthreshold = threshold;
options.omax = octaves;
options.nsublevels = sublevels;
options.diffusivity = diffusivity;
AKAZEFeatures impl(options);
impl.Create_Nonlinear_Scale_Space(img1_32);
@ -188,7 +204,7 @@ namespace cv
img.convertTo(img1_32, CV_32F, 1.0 / 255.0, 0);
AKAZEOptions options;
options.descriptor = static_cast<DESCRIPTOR_TYPE>(descriptor);
options.descriptor = descriptor;
options.descriptor_channels = descriptor_channels;
options.descriptor_size = descriptor_size;
options.img_width = img.cols;
@ -216,7 +232,7 @@ namespace cv
cv::Mat& desc = descriptors.getMatRef();
AKAZEOptions options;
options.descriptor = static_cast<DESCRIPTOR_TYPE>(descriptor);
options.descriptor = descriptor;
options.descriptor_channels = descriptor_channels;
options.descriptor_size = descriptor_size;
options.img_width = img.cols;
@ -229,4 +245,4 @@ namespace cv
CV_Assert((!desc.rows || desc.cols == descriptorSize()));
CV_Assert((!desc.rows || (desc.type() == descriptorType())));
}
}
}

1941
modules/features2d/src/akaze/AKAZEFeatures.cpp
View File

File diff suppressed because it is too large Load Diff

65
modules/features2d/src/akaze/AKAZEFeatures.h
View File

@ -1,65 +0,0 @@
/**
* @file AKAZE.h
* @brief Main class for detecting and computing binary descriptors in an
* accelerated nonlinear scale space
* @date Mar 27, 2013
* @author Pablo F. Alcantarilla, Jesus Nuevo
*/
#pragma once
/* ************************************************************************* */
// Includes
#include "precomp.hpp"
#include "AKAZEConfig.h"
/* ************************************************************************* */
// AKAZE Class Declaration
class AKAZEFeatures {
private:
AKAZEOptions options_; ///< Configuration options for AKAZE
std::vector<TEvolution> evolution_; ///< Vector of nonlinear diffusion evolution
/// FED parameters
int ncycles_; ///< Number of cycles
bool reordering_; ///< Flag for reordering time steps
std::vector<std::vector<float > > tsteps_; ///< Vector of FED dynamic time steps
std::vector<int> nsteps_; ///< Vector of number of steps per cycle
/// Matrices for the M-LDB descriptor computation
cv::Mat descriptorSamples_; // List of positions in the grids to sample LDB bits from.
cv::Mat descriptorBits_;
cv::Mat bitMask_;
public:
/// Constructor with input arguments
AKAZEFeatures(const AKAZEOptions& options);
/// Scale Space methods
void Allocate_Memory_Evolution();
int Create_Nonlinear_Scale_Space(const cv::Mat& img);
void Feature_Detection(std::vector<cv::KeyPoint>& kpts);
void Compute_Determinant_Hessian_Response(void);
void Compute_Multiscale_Derivatives(void);
void Find_Scale_Space_Extrema(std::vector<cv::KeyPoint>& kpts);
void Do_Subpixel_Refinement(std::vector<cv::KeyPoint>& kpts);
// Feature description methods
void Compute_Descriptors(std::vector<cv::KeyPoint>& kpts, cv::Mat& desc);
static void Compute_Main_Orientation(cv::KeyPoint& kpt, const std::vector<TEvolution>& evolution_);
};
/* ************************************************************************* */
// Inline functions
// Inline functions
void generateDescriptorSubsample(cv::Mat& sampleList, cv::Mat& comparisons,
int nbits, int pattern_size, int nchannels);
float get_angle(float x, float y);
float gaussian(float x, float y, float sigma);
void check_descriptor_limits(int& x, int& y, int width, int height);
int fRound(float flt);

17
modules/features2d/src/kaze.cpp
View File

@ -55,12 +55,21 @@ namespace cv
KAZE::KAZE()
: extended(false)
, upright(false)
, threshold(0.001f)
, octaves(4)
, sublevels(4)
, diffusivity(DIFF_PM_G2)
{
}
KAZE::KAZE(bool _extended, bool _upright)
KAZE::KAZE(bool _extended, bool _upright, float _threshold, int _octaves,
int _sublevels, int _diffusivity)
: extended(_extended)
, upright(_upright)
, threshold(_threshold)
, octaves(_octaves)
, sublevels(_sublevels)
, diffusivity(_diffusivity)
{
}
@ -111,6 +120,10 @@ namespace cv
options.img_height = img.rows;
options.extended = extended;
options.upright = upright;
options.dthreshold = threshold;
options.omax = octaves;
options.nsublevels = sublevels;
options.diffusivity = diffusivity;
KAZEFeatures impl(options);
impl.Create_Nonlinear_Scale_Space(img1_32);
@ -180,4 +193,4 @@ namespace cv
CV_Assert((!desc.rows || desc.cols == descriptorSize()));
CV_Assert((!desc.rows || (desc.type() == descriptorType())));
}
}
}

45
modules/features2d/src/akaze/AKAZEConfig.h → modules/features2d/src/kaze/AKAZEConfig.h
View File

@ -5,7 +5,8 @@
* @author Pablo F. Alcantarilla, Jesus Nuevo
*/
#pragma once
#ifndef __OPENCV_FEATURES_2D_AKAZE_CONFIG_H__
#define __OPENCV_FEATURES_2D_AKAZE_CONFIG_H__
/* ************************************************************************* */
// OpenCV
@ -28,14 +29,6 @@ const float gauss25[7][7] = {
/// AKAZE configuration options structure
struct AKAZEOptions {
/// AKAZE Diffusivities
enum DIFFUSIVITY_TYPE {
PM_G1 = 0,
PM_G2 = 1,
WEICKERT = 2,
CHARBONNIER = 3
};
AKAZEOptions()
: omax(4)
, nsublevels(4)
@ -44,12 +37,12 @@ struct AKAZEOptions {
, soffset(1.6f)
, derivative_factor(1.5f)
, sderivatives(1.0)
, diffusivity(PM_G2)
, diffusivity(cv::DIFF_PM_G2)
, dthreshold(0.001f)
, min_dthreshold(0.00001f)
, descriptor(cv::AKAZE::DESCRIPTOR_MLDB)
, descriptor(cv::DESCRIPTOR_MLDB)
, descriptor_size(0)
, descriptor_channels(3)
, descriptor_pattern_size(10)
@ -67,12 +60,12 @@ struct AKAZEOptions {
float soffset; ///< Base scale offset (sigma units)
float derivative_factor; ///< Factor for the multiscale derivatives
float sderivatives; ///< Smoothing factor for the derivatives
DIFFUSIVITY_TYPE diffusivity; ///< Diffusivity type
int diffusivity; ///< Diffusivity type
float dthreshold; ///< Detector response threshold to accept point
float min_dthreshold; ///< Minimum detector threshold to accept a point
cv::AKAZE::DESCRIPTOR_TYPE descriptor; ///< Type of descriptor
int descriptor; ///< Type of descriptor
int descriptor_size; ///< Size of the descriptor in bits. 0->Full size
int descriptor_channels; ///< Number of channels in the descriptor (1, 2, 3)
int descriptor_pattern_size; ///< Actual patch size is 2*pattern_size*point.scale
@ -82,28 +75,4 @@ struct AKAZEOptions {
int kcontrast_nbins; ///< Number of bins for the contrast factor histogram
};
/* ************************************************************************* */
/// AKAZE nonlinear diffusion filtering evolution
struct TEvolution {
TEvolution() {
etime = 0.0f;
esigma = 0.0f;
octave = 0;
sublevel = 0;
sigma_size = 0;
}
cv::Mat Lx, Ly; // First order spatial derivatives
cv::Mat Lxx, Lxy, Lyy; // Second order spatial derivatives
cv::Mat Lflow; // Diffusivity image
cv::Mat Lt; // Evolution image
cv::Mat Lsmooth; // Smoothed image
cv::Mat Lstep; // Evolution step update
cv::Mat Ldet; // Detector response
float etime; // Evolution time
float esigma; // Evolution sigma. For linear diffusion t = sigma^2 / 2
size_t octave; // Image octave
size_t sublevel; // Image sublevel in each octave
size_t sigma_size; // Integer sigma. For computing the feature detector responses
};
#endif

1880
modules/features2d/src/kaze/AKAZEFeatures.cpp Normal file
View File

File diff suppressed because it is too large Load Diff

62
modules/features2d/src/kaze/AKAZEFeatures.h Normal file
View File

@ -0,0 +1,62 @@
/**
* @file AKAZE.h
* @brief Main class for detecting and computing binary descriptors in an
* accelerated nonlinear scale space
* @date Mar 27, 2013
* @author Pablo F. Alcantarilla, Jesus Nuevo
*/
#ifndef __OPENCV_FEATURES_2D_AKAZE_FEATURES_H__
#define __OPENCV_FEATURES_2D_AKAZE_FEATURES_H__
/* ************************************************************************* */
// Includes
#include "precomp.hpp"
#include "AKAZEConfig.h"
#include "TEvolution.h"
/* ************************************************************************* */
// AKAZE Class Declaration
class AKAZEFeatures {
private:
AKAZEOptions options_; ///< Configuration options for AKAZE
std::vector<TEvolution> evolution_; ///< Vector of nonlinear diffusion evolution
/// FED parameters
int ncycles_; ///< Number of cycles
bool reordering_; ///< Flag for reordering time steps
std::vector<std::vector<float > > tsteps_; ///< Vector of FED dynamic time steps
std::vector<int> nsteps_; ///< Vector of number of steps per cycle
/// Matrices for the M-LDB descriptor computation
cv::Mat descriptorSamples_; // List of positions in the grids to sample LDB bits from.
cv::Mat descriptorBits_;
cv::Mat bitMask_;
public:
/// Constructor with input arguments
AKAZEFeatures(const AKAZEOptions& options);
/// Scale Space methods
void Allocate_Memory_Evolution();
int Create_Nonlinear_Scale_Space(const cv::Mat& img);
void Feature_Detection(std::vector<cv::KeyPoint>& kpts);
void Compute_Determinant_Hessian_Response(void);
void Compute_Multiscale_Derivatives(void);
void Find_Scale_Space_Extrema(std::vector<cv::KeyPoint>& kpts);
void Do_Subpixel_Refinement(std::vector<cv::KeyPoint>& kpts);
/// Feature description methods
void Compute_Descriptors(std::vector<cv::KeyPoint>& kpts, cv::Mat& desc);
static void Compute_Main_Orientation(cv::KeyPoint& kpt, const std::vector<TEvolution>& evolution_);
};
/* ************************************************************************* */
/// Inline functions
void generateDescriptorSubsample(cv::Mat& sampleList, cv::Mat& comparisons,
int nbits, int pattern_size, int nchannels);
#endif

43
modules/features2d/src/kaze/KAZEConfig.h
View File

@ -5,7 +5,8 @@
* @author Pablo F. Alcantarilla
*/
#pragma once
#ifndef __OPENCV_FEATURES_2D_AKAZE_CONFIG_H__
#define __OPENCV_FEATURES_2D_AKAZE_CONFIG_H__
// OpenCV Includes
#include "precomp.hpp"
@ -15,14 +16,8 @@
struct KAZEOptions {
enum DIFFUSIVITY_TYPE {
PM_G1 = 0,
PM_G2 = 1,
WEICKERT = 2
};
KAZEOptions()
: diffusivity(PM_G2)
: diffusivity(cv::DIFF_PM_G2)
, soffset(1.60f)
, omax(4)
@ -33,20 +28,13 @@ struct KAZEOptions {
, dthreshold(0.001f)
, kcontrast(0.01f)
, kcontrast_percentille(0.7f)
, kcontrast_bins(300)
, use_fed(true)
, kcontrast_bins(300)
, upright(false)
, extended(false)
, use_clipping_normalilzation(false)
, clipping_normalization_ratio(1.6f)
, clipping_normalization_niter(5)
{
}
DIFFUSIVITY_TYPE diffusivity;
int diffusivity;
float soffset;
int omax;
int nsublevels;
@ -57,27 +45,8 @@ struct KAZEOptions {
float kcontrast;
float kcontrast_percentille;
int kcontrast_bins;
bool use_fed;
bool upright;
bool extended;
bool use_clipping_normalilzation;
float clipping_normalization_ratio;
int clipping_normalization_niter;
};
struct TEvolution {
cv::Mat Lx, Ly; // First order spatial derivatives
cv::Mat Lxx, Lxy, Lyy; // Second order spatial derivatives
cv::Mat Lflow; // Diffusivity image
cv::Mat Lt; // Evolution image
cv::Mat Lsmooth; // Smoothed image
cv::Mat Lstep; // Evolution step update
cv::Mat Ldet; // Detector response
float etime; // Evolution time
float esigma; // Evolution sigma. For linear diffusion t = sigma^2 / 2
float octave; // Image octave
float sublevel; // Image sublevel in each octave
int sigma_size; // Integer esigma. For computing the feature detector responses
};
#endif

808
modules/features2d/src/kaze/KAZEFeatures.cpp
View File

File diff suppressed because it is too large Load Diff

69
modules/features2d/src/kaze/KAZEFeatures.h
View File

@ -7,84 +7,53 @@
* @author Pablo F. Alcantarilla
*/
#ifndef KAZE_H_
#define KAZE_H_
//*************************************************************************************
//*************************************************************************************
#ifndef __OPENCV_FEATURES_2D_KAZE_FEATURES_H__
#define __OPENCV_FEATURES_2D_KAZE_FEATURES_H__
/* ************************************************************************* */
// Includes
#include "KAZEConfig.h"
#include "nldiffusion_functions.h"
#include "fed.h"
#include "TEvolution.h"
//*************************************************************************************
//*************************************************************************************
/* ************************************************************************* */
// KAZE Class Declaration
class KAZEFeatures {
private:
KAZEOptions options;
/// Parameters of the Nonlinear diffusion class
KAZEOptions options_; ///< Configuration options for KAZE
std::vector<TEvolution> evolution_; ///< Vector of nonlinear diffusion evolution
// Parameters of the Nonlinear diffusion class
std::vector<TEvolution> evolution_; // Vector of nonlinear diffusion evolution
// Vector of keypoint vectors for finding extrema in multiple threads
/// Vector of keypoint vectors for finding extrema in multiple threads
std::vector<std::vector<cv::KeyPoint> > kpts_par_;
// FED parameters
int ncycles_; // Number of cycles
bool reordering_; // Flag for reordering time steps
std::vector<std::vector<float > > tsteps_; // Vector of FED dynamic time steps
std::vector<int> nsteps_; // Vector of number of steps per cycle
// Some auxiliary variables used in the AOS step
cv::Mat Ltx_, Lty_, px_, py_, ax_, ay_, bx_, by_, qr_, qc_;
/// FED parameters
int ncycles_; ///< Number of cycles
bool reordering_; ///< Flag for reordering time steps
std::vector<std::vector<float > > tsteps_; ///< Vector of FED dynamic time steps
std::vector<int> nsteps_; ///< Vector of number of steps per cycle
public:
// Constructor
/// Constructor
KAZEFeatures(KAZEOptions& options);
// Public methods for KAZE interface
/// Public methods for KAZE interface
void Allocate_Memory_Evolution(void);
int Create_Nonlinear_Scale_Space(const cv::Mat& img);
void Feature_Detection(std::vector<cv::KeyPoint>& kpts);
void Feature_Description(std::vector<cv::KeyPoint>& kpts, cv::Mat& desc);
static void Compute_Main_Orientation(cv::KeyPoint& kpt, const std::vector<TEvolution>& evolution_, const KAZEOptions& options);
private:
// Feature Detection Methods
/// Feature Detection Methods
void Compute_KContrast(const cv::Mat& img, const float& kper);
void Compute_Multiscale_Derivatives(void);
void Compute_Detector_Response(void);
void Determinant_Hessian_Parallel(std::vector<cv::KeyPoint>& kpts);
void Find_Extremum_Threading(const int& level);
void Determinant_Hessian(std::vector<cv::KeyPoint>& kpts);
void Do_Subpixel_Refinement(std::vector<cv::KeyPoint>& kpts);
// AOS Methods
void AOS_Step_Scalar(cv::Mat &Ld, const cv::Mat &Ldprev, const cv::Mat &c, const float& stepsize);
void AOS_Rows(const cv::Mat &Ldprev, const cv::Mat &c, const float& stepsize);
void AOS_Columns(const cv::Mat &Ldprev, const cv::Mat &c, const float& stepsize);
void Thomas(const cv::Mat &a, const cv::Mat &b, const cv::Mat &Ld, cv::Mat &x);
};
//*************************************************************************************
//*************************************************************************************
// Inline functions
float getAngle(const float& x, const float& y);
float gaussian(const float& x, const float& y, const float& sig);
void checkDescriptorLimits(int &x, int &y, const int& width, const int& height);
void clippingDescriptor(float *desc, const int& dsize, const int& niter, const float& ratio);
int fRound(const float& flt);
//*************************************************************************************
//*************************************************************************************
#endif // KAZE_H_
#endif

35
modules/features2d/src/kaze/TEvolution.h Normal file
View File

@ -0,0 +1,35 @@
/**
* @file TEvolution.h
* @brief Header file with the declaration of the TEvolution struct
* @date Jun 02, 2014
* @author Pablo F. Alcantarilla
*/
#ifndef __OPENCV_FEATURES_2D_TEVOLUTION_H__
#define __OPENCV_FEATURES_2D_TEVOLUTION_H__
/* ************************************************************************* */
/// KAZE/A-KAZE nonlinear diffusion filtering evolution
struct TEvolution {
TEvolution() {
etime = 0.0f;
esigma = 0.0f;
octave = 0;
sublevel = 0;
sigma_size = 0;
}
cv::Mat Lx, Ly; ///< First order spatial derivatives
cv::Mat Lxx, Lxy, Lyy; ///< Second order spatial derivatives
cv::Mat Lt; ///< Evolution image
cv::Mat Lsmooth; ///< Smoothed image
cv::Mat Ldet; ///< Detector response
float etime; ///< Evolution time
float esigma; ///< Evolution sigma. For linear diffusion t = sigma^2 / 2
int octave; ///< Image octave
int sublevel; ///< Image sublevel in each octave
int sigma_size; ///< Integer esigma. For computing the feature detector responses
};
#endif

6
modules/features2d/src/kaze/fed.h
View File

@ -1,5 +1,5 @@
#ifndef FED_H
#define FED_H
#ifndef __OPENCV_FEATURES_2D_FED_H__
#define __OPENCV_FEATURES_2D_FED_H__
//******************************************************************************
//******************************************************************************
@ -22,4 +22,4 @@ bool fed_is_prime_internal(const int& number);
//*************************************************************************************
//*************************************************************************************
#endif // FED_H
#endif // __OPENCV_FEATURES_2D_FED_H__

4
modules/features2d/src/kaze/nldiffusion_functions.h
View File

@ -8,8 +8,8 @@
* @author Pablo F. Alcantarilla
*/
#ifndef KAZE_NLDIFFUSION_FUNCTIONS_H
#define KAZE_NLDIFFUSION_FUNCTIONS_H
#ifndef __OPENCV_FEATURES_2D_NLDIFFUSION_FUNCTIONS_H__
#define __OPENCV_FEATURES_2D_NLDIFFUSION_FUNCTIONS_H__
/* ************************************************************************* */
// Includes

77
modules/features2d/src/kaze/utils.h Normal file
View File

@ -0,0 +1,77 @@
#ifndef __OPENCV_FEATURES_2D_KAZE_UTILS_H__
#define __OPENCV_FEATURES_2D_KAZE_UTILS_H__
/* ************************************************************************* */
/**
* @brief This function computes the angle from the vector given by (X Y). From 0 to 2*Pi
*/
inline float getAngle(float x, float y) {
if (x >= 0 && y >= 0) {
return atanf(y / x);
}
if (x < 0 && y >= 0) {
return static_cast<float>(CV_PI)-atanf(-y / x);
}
if (x < 0 && y < 0) {
return static_cast<float>(CV_PI)+atanf(y / x);
}
if (x >= 0 && y < 0) {
return static_cast<float>(2.0 * CV_PI) - atanf(-y / x);
}
return 0;
}
/* ************************************************************************* */
/**
* @brief This function computes the value of a 2D Gaussian function
* @param x X Position
* @param y Y Position
* @param sig Standard Deviation
*/
inline float gaussian(float x, float y, float sigma) {
return expf(-(x*x + y*y) / (2.0f*sigma*sigma));
}
/* ************************************************************************* */
/**
* @brief This function checks descriptor limits
* @param x X Position
* @param y Y Position
* @param width Image width
* @param height Image height
*/
inline void checkDescriptorLimits(int &x, int &y, int width, int height) {
if (x < 0) {
x = 0;
}
if (y < 0) {
y = 0;
}
if (x > width - 1) {
x = width - 1;
}
if (y > height - 1) {
y = height - 1;
}
}
/* ************************************************************************* */
/**
* @brief This funtion rounds float to nearest integer
* @param flt Input float
* @return dst Nearest integer
*/
inline int fRound(float flt) {
return (int)(flt + 0.5f);
}
#endif

40
modules/features2d/test/test_descriptors_regression.cpp
View File

@ -101,8 +101,14 @@ public:
typedef typename Distance::ResultType DistanceType;
CV_DescriptorExtractorTest( const string _name, DistanceType _maxDist, const Ptr<DescriptorExtractor>& _dextractor,
Distance d = Distance() ):
name(_name), maxDist(_maxDist), dextractor(_dextractor), distance(d) {}
Distance d = Distance(), Ptr<FeatureDetector> _detector = Ptr<FeatureDetector>()):
name(_name), maxDist(_maxDist), dextractor(_dextractor), distance(d) , detector(_detector) {}
~CV_DescriptorExtractorTest()
{
if(!detector.empty())
detector.release();
}
protected:
virtual void createDescriptorExtractor() {}
@ -189,7 +195,6 @@ protected:
// Read the test image.
string imgFilename = string(ts->get_data_path()) + FEATURES2D_DIR + "/" + IMAGE_FILENAME;
Mat img = imread( imgFilename );
if( img.empty() )
{
@ -197,13 +202,15 @@ protected:
ts->set_failed_test_info( cvtest::TS::FAIL_INVALID_TEST_DATA );
return;
}
vector<KeyPoint> keypoints;
FileStorage fs( string(ts->get_data_path()) + FEATURES2D_DIR + "/keypoints.xml.gz", FileStorage::READ );
if( fs.isOpened() )
{
if(!detector.empty()) {
detector->detect(img, keypoints);
} else {
read( fs.getFirstTopLevelNode(), keypoints );
}
if(!keypoints.empty())
{
Mat calcDescriptors;
double t = (double)getTickCount();
dextractor->compute( img, keypoints, calcDescriptors );
@ -244,7 +251,7 @@ protected:
}
}
}
else
if(!fs.isOpened())
{
ts->printf( cvtest::TS::LOG, "Compute and write keypoints.\n" );
fs.open( string(ts->get_data_path()) + FEATURES2D_DIR + "/keypoints.xml.gz", FileStorage::WRITE );
@ -295,6 +302,7 @@ protected:
const DistanceType maxDist;
Ptr<DescriptorExtractor> dextractor;
Distance distance;
Ptr<FeatureDetector> detector;
private:
CV_DescriptorExtractorTest& operator=(const CV_DescriptorExtractorTest&) { return *this; }
@ -340,3 +348,19 @@ TEST( Features2d_DescriptorExtractor_OpponentBRIEF, regression )
DescriptorExtractor::create("OpponentBRIEF") );
test.safe_run();
}
TEST( Features2d_DescriptorExtractor_KAZE, regression )
{
CV_DescriptorExtractorTest< L2<float> > test( "descriptor-kaze", 0.03f,
DescriptorExtractor::create("KAZE"),
L2<float>(), FeatureDetector::create("KAZE"));
test.safe_run();
}
TEST( Features2d_DescriptorExtractor_AKAZE, regression )
{
CV_DescriptorExtractorTest<Hamming> test( "descriptor-akaze", (CV_DescriptorExtractorTest<Hamming>::DistanceType)12.f,
DescriptorExtractor::create("AKAZE"),
Hamming(), FeatureDetector::create("AKAZE"));
test.safe_run();
}

12
modules/features2d/test/test_detectors_regression.cpp
View File

@ -289,6 +289,18 @@ TEST( Features2d_Detector_ORB, regression )
test.safe_run();
}
TEST( Features2d_Detector_KAZE, regression )
{
CV_FeatureDetectorTest test( "detector-kaze", FeatureDetector::create("KAZE") );
test.safe_run();
}
TEST( Features2d_Detector_AKAZE, regression )
{
CV_FeatureDetectorTest test( "detector-akaze", FeatureDetector::create("AKAZE") );
test.safe_run();
}
TEST( Features2d_Detector_GridFAST, regression )
{
CV_FeatureDetectorTest test( "detector-grid-fast", FeatureDetector::create("GridFAST") );

10
modules/features2d/test/test_keypoints.cpp
View File

@ -167,19 +167,17 @@ TEST(Features2d_Detector_Keypoints_Dense, validation)
test.safe_run();
}
// FIXIT #2807 Crash on Windows 7 x64 MSVS 2012, Linux Fedora 19 x64 with GCC 4.8.2, Linux Ubuntu 14.04 LTS x64 with GCC 4.8.2
TEST(Features2d_Detector_Keypoints_KAZE, DISABLED_validation)
TEST(Features2d_Detector_Keypoints_KAZE, validation)
{
CV_FeatureDetectorKeypointsTest test(Algorithm::create<FeatureDetector>("Feature2D.KAZE"));
test.safe_run();
}
// FIXIT #2807 Crash on Windows 7 x64 MSVS 2012, Linux Fedora 19 x64 with GCC 4.8.2, Linux Ubuntu 14.04 LTS x64 with GCC 4.8.2
TEST(Features2d_Detector_Keypoints_AKAZE, DISABLED_validation)
TEST(Features2d_Detector_Keypoints_AKAZE, validation)
{
CV_FeatureDetectorKeypointsTest test_kaze(cv::Ptr<FeatureDetector>(new cv::AKAZE(cv::AKAZE::DESCRIPTOR_KAZE)));
CV_FeatureDetectorKeypointsTest test_kaze(cv::Ptr<FeatureDetector>(new cv::AKAZE(cv::DESCRIPTOR_KAZE)));
test_kaze.safe_run();
CV_FeatureDetectorKeypointsTest test_mldb(cv::Ptr<FeatureDetector>(new cv::AKAZE(cv::AKAZE::DESCRIPTOR_MLDB)));
CV_FeatureDetectorKeypointsTest test_mldb(cv::Ptr<FeatureDetector>(new cv::AKAZE(cv::DESCRIPTOR_MLDB)));
test_mldb.safe_run();
}

6
modules/features2d/test/test_rotation_and_scale_invariance.cpp
View File

@ -652,8 +652,7 @@ TEST(Features2d_ScaleInvariance_Detector_BRISK, regression)
test.safe_run();
}
// FIXIT #2807 Crash on Windows 7 x64 MSVS 2012, Linux Fedora 19 x64 with GCC 4.8.2, Linux Ubuntu 14.04 LTS x64 with GCC 4.8.2
TEST(Features2d_ScaleInvariance_Detector_KAZE, DISABLED_regression)
TEST(Features2d_ScaleInvariance_Detector_KAZE, regression)
{
DetectorScaleInvarianceTest test(Algorithm::create<FeatureDetector>("Feature2D.KAZE"),
0.08f,
@ -661,8 +660,7 @@ TEST(Features2d_ScaleInvariance_Detector_KAZE, DISABLED_regression)
test.safe_run();
}
// FIXIT #2807 Crash on Windows 7 x64 MSVS 2012, Linux Fedora 19 x64 with GCC 4.8.2, Linux Ubuntu 14.04 LTS x64 with GCC 4.8.2
TEST(Features2d_ScaleInvariance_Detector_AKAZE, DISABLED_regression)
TEST(Features2d_ScaleInvariance_Detector_AKAZE, regression)
{
DetectorScaleInvarianceTest test(Algorithm::create<FeatureDetector>("Feature2D.AKAZE"),
0.08f,

3
modules/highgui/doc/user_interface.rst
View File

@ -83,6 +83,9 @@ If window was created with OpenGL support, ``imshow`` also support :ocv:class:`o
.. note:: This function should be followed by ``waitKey`` function which displays the image for specified milliseconds. Otherwise, it won't display the image. For example, ``waitKey(0)`` will display the window infinitely until any keypress (it is suitable for image display). ``waitKey(25)`` will display a frame for 25 ms, after which display will be automatically closed. (If you put it in a loop to read videos, it will display the video frame-by-frame)
.. note::
[Windows Backend Only] Pressing Ctrl+C will copy the image to the clipboard.
namedWindow
---------------

84
modules/highgui/src/window_w32.cpp
View File

@ -1286,6 +1286,10 @@ MainWindowProc( HWND hwnd, UINT uMsg, WPARAM wParam, LPARAM lParam )
switch(uMsg)
{
case WM_COPY:
::WindowProc(hwnd, uMsg, wParam, lParam); // call highgui proc. There may be a better way to do this.
break;
case WM_DESTROY:
icvRemoveWindow(window);
@ -1448,6 +1452,81 @@ static LRESULT CALLBACK HighGUIProc( HWND hwnd, UINT uMsg, WPARAM wParam, LPARAM
// Process the message
switch(uMsg)
{
case WM_COPY:
{
if (!::OpenClipboard(hwnd) )
break;
HDC hDC = 0;
HDC memDC = 0;
HBITMAP memBM = 0;
// We'll use a do-while(0){} scope as a single-run breakable scope
// Upon any error we can jump out of the single-time while scope to clean up the resources.
do
{
if (!::EmptyClipboard())
break;
if(!window->image)
break;
// Get window device context
if (0 == (hDC = ::GetDC(hwnd)))
break;
// Create another DC compatible with hDC
if (0 == (memDC = ::CreateCompatibleDC( hDC )))
break;
// Determine the bitmap's dimensions
int nchannels = 3;
SIZE size = {0,0};
icvGetBitmapData( window, &size, &nchannels, 0 );
// Create bitmap to draw on and it in the new DC
if (0 == (memBM = ::CreateCompatibleBitmap ( hDC, size.cx, size.cy)))
break;
if (!::SelectObject( memDC, memBM ))
break;
// Begin drawing to DC
if (!::SetStretchBltMode(memDC, COLORONCOLOR))
break;
RGBQUAD table[256];
if( 1 == nchannels )
{
for(int i = 0; i < 256; ++i)
{
table[i].rgbBlue = (unsigned char)i;
table[i].rgbGreen = (unsigned char)i;
table[i].rgbRed = (unsigned char)i;
}
if (!::SetDIBColorTable(window->dc, 0, 255, table))
break;
}
// The image copied to the clipboard will be in its original size, regardless if the window itself was resized.
// Render the image to the dc/bitmap (at original size).
if (!::BitBlt( memDC, 0, 0, size.cx, size.cy, window->dc, 0, 0, SRCCOPY ))
break;
// Finally, set bitmap to clipboard
::SetClipboardData(CF_BITMAP, memBM);
} while (0,0); // (0,0) instead of (0) to avoid MSVC compiler warning C4127: "conditional expression is constant"
//////////////////////////////////////////////////////////////////////////
// if handle is allocated (i.e. != 0) then clean-up.
if (memBM) ::DeleteObject(memBM);
if (memDC) ::DeleteDC(memDC);
if (hDC) ::ReleaseDC(hwnd, hDC);
::CloseClipboard();
break;
}
case WM_WINDOWPOSCHANGING:
{
LPWINDOWPOS pos = (LPWINDOWPOS)lParam;
@ -1798,6 +1877,11 @@ cvWaitKey( int delay )
is_processed = 1;
return (int)(message.wParam << 16);
}
// Intercept Ctrl+C for copy to clipboard
if ('C' == message.wParam && (::GetKeyState(VK_CONTROL)>>15))
::PostMessage(message.hwnd, WM_COPY, 0, 0);
default:
DispatchMessage(&message);
is_processed = 1;

19
modules/imgcodecs/include/opencv2/imgcodecs.hpp
View File

@ -56,14 +56,17 @@ enum { IMREAD_UNCHANGED = -1, // 8bit, color or not
IMREAD_ANYCOLOR = 4 // ?, any color
};
enum { IMWRITE_JPEG_QUALITY = 1,
IMWRITE_JPEG_PROGRESSIVE = 2,
IMWRITE_JPEG_OPTIMIZE = 3,
IMWRITE_PNG_COMPRESSION = 16,
IMWRITE_PNG_STRATEGY = 17,
IMWRITE_PNG_BILEVEL = 18,
IMWRITE_PXM_BINARY = 32,
IMWRITE_WEBP_QUALITY = 64
enum { IMWRITE_JPEG_QUALITY = 1,
IMWRITE_JPEG_PROGRESSIVE = 2,
IMWRITE_JPEG_OPTIMIZE = 3,
IMWRITE_JPEG_RST_INTERVAL = 4,
IMWRITE_JPEG_LUMA_QUALITY = 5,
IMWRITE_JPEG_CHROMA_QUALITY = 6,
IMWRITE_PNG_COMPRESSION = 16,
IMWRITE_PNG_STRATEGY = 17,
IMWRITE_PNG_BILEVEL = 18,
IMWRITE_PXM_BINARY = 32,
IMWRITE_WEBP_QUALITY = 64
};
enum { IMWRITE_PNG_STRATEGY_DEFAULT = 0,

3
modules/imgcodecs/include/opencv2/imgcodecs/imgcodecs_c.h
View File

@ -76,6 +76,9 @@ enum
CV_IMWRITE_JPEG_QUALITY =1,
CV_IMWRITE_JPEG_PROGRESSIVE =2,
CV_IMWRITE_JPEG_OPTIMIZE =3,
CV_IMWRITE_JPEG_RST_INTERVAL =4,
CV_IMWRITE_JPEG_LUMA_QUALITY =5,
CV_IMWRITE_JPEG_CHROMA_QUALITY =6,
CV_IMWRITE_PNG_COMPRESSION =16,
CV_IMWRITE_PNG_STRATEGY =17,
CV_IMWRITE_PNG_BILEVEL =18,

52
modules/imgcodecs/src/grfmt_jpeg.cpp
View File

@ -600,6 +600,9 @@ bool JpegEncoder::write( const Mat& img, const std::vector<int>& params )
int quality = 95;
int progressive = 0;
int optimize = 0;
int rst_interval = 0;
int luma_quality = -1;
int chroma_quality = -1;
for( size_t i = 0; i < params.size(); i += 2 )
{
@ -618,15 +621,64 @@ bool JpegEncoder::write( const Mat& img, const std::vector<int>& params )
{
optimize = params[i+1];
}
if( params[i] == CV_IMWRITE_JPEG_LUMA_QUALITY )
{
if (params[i+1] >= 0)
{
luma_quality = MIN(MAX(params[i+1], 0), 100);
quality = luma_quality;
if (chroma_quality < 0)
{
chroma_quality = luma_quality;
}
}
}
if( params[i] == CV_IMWRITE_JPEG_CHROMA_QUALITY )
{
if (params[i+1] >= 0)
{
chroma_quality = MIN(MAX(params[i+1], 0), 100);
}
}
if( params[i] == CV_IMWRITE_JPEG_RST_INTERVAL )
{
rst_interval = params[i+1];
rst_interval = MIN(MAX(rst_interval, 0), 65535L);
}
}
jpeg_set_defaults( &cinfo );
cinfo.restart_interval = rst_interval;
jpeg_set_quality( &cinfo, quality,
TRUE /* limit to baseline-JPEG values */ );
if( progressive )
jpeg_simple_progression( &cinfo );
if( optimize )
cinfo.optimize_coding = TRUE;
#if JPEG_LIB_VERSION >= 70
if (luma_quality >= 0 && chroma_quality >= 0)
{
cinfo.q_scale_factor[0] = jpeg_quality_scaling(luma_quality);
cinfo.q_scale_factor[1] = jpeg_quality_scaling(chroma_quality);
if ( luma_quality != chroma_quality )
{
/* disable subsampling - ref. Libjpeg.txt */
cinfo.comp_info[0].v_samp_factor = 1;
cinfo.comp_info[0].h_samp_factor = 1;
cinfo.comp_info[1].v_samp_factor = 1;
cinfo.comp_info[1].h_samp_factor = 1;
}
jpeg_default_qtables( &cinfo, TRUE );
}
#endif // #if JPEG_LIB_VERSION >= 70
jpeg_start_compress( &cinfo, TRUE );
if( channels > 1 )

17
modules/imgcodecs/src/grfmt_tiff.cpp
View File

@ -158,7 +158,7 @@ bool TiffDecoder::readHeader()
m_type = CV_MAKETYPE(CV_8U, photometric > 1 ? wanted_channels : 1);
break;
case 16:
m_type = CV_MAKETYPE(CV_16U, photometric > 1 ? 3 : 1);
m_type = CV_MAKETYPE(CV_16U, photometric > 1 ? wanted_channels : 1);
break;
case 32:
@ -326,6 +326,21 @@ bool TiffDecoder::readData( Mat& img )
(ushort*)(data + img.step*i) + x*3, 0,
cvSize(tile_width,1) );
}
else if (ncn == 4)
{
if (wanted_channels == 4)
{
icvCvt_BGRA2RGBA_16u_C4R(buffer16 + i*tile_width0*ncn, 0,
(ushort*)(data + img.step*i) + x * 4, 0,
cvSize(tile_width, 1));
}
else
{
icvCvt_BGRA2BGR_16u_C4C3R(buffer16 + i*tile_width0*ncn, 0,
(ushort*)(data + img.step*i) + x * 3, 0,
cvSize(tile_width, 1), 2);
}
}
else
{
icvCvt_BGRA2BGR_16u_C4C3R(buffer16 + i*tile_width0*ncn, 0,

31
modules/imgcodecs/test/test_grfmt.cpp
View File

@ -139,9 +139,6 @@ public:
string filename = cv::tempfile(".jpg");
imwrite(filename, img);
img = imread(filename, IMREAD_UNCHANGED);
filename = string(ts->get_data_path() + "readwrite/test_" + char(k + 48) + "_c" + char(num_channels + 48) + ".jpg");
ts->printf(ts->LOG, "reading test image : %s\n", filename.c_str());
Mat img_test = imread(filename, IMREAD_UNCHANGED);
@ -160,8 +157,9 @@ public:
#endif
#ifdef HAVE_TIFF
for (int num_channels = 1; num_channels <= 3; num_channels+=2)
for (int num_channels = 1; num_channels <= 4; num_channels++)
{
if (num_channels == 2) continue;
// tiff
ts->printf(ts->LOG, "image type depth:%d channels:%d ext: %s\n", CV_16U, num_channels, ".tiff");
Mat img(img_r * k, img_c * k, CV_MAKETYPE(CV_16U, num_channels), Scalar::all(0));
@ -433,6 +431,31 @@ TEST(Imgcodecs_Jpeg, encode_decode_optimize_jpeg)
remove(output_optimized.c_str());
}
TEST(Imgcodecs_Jpeg, encode_decode_rst_jpeg)
{
cvtest::TS& ts = *cvtest::TS::ptr();
string input = string(ts.get_data_path()) + "../cv/shared/lena.png";
cv::Mat img = cv::imread(input);
ASSERT_FALSE(img.empty());
std::vector<int> params;
params.push_back(IMWRITE_JPEG_RST_INTERVAL);
params.push_back(1);
string output_rst = cv::tempfile(".jpg");
EXPECT_NO_THROW(cv::imwrite(output_rst, img, params));
cv::Mat img_jpg_rst = cv::imread(output_rst);
string output_normal = cv::tempfile(".jpg");
EXPECT_NO_THROW(cv::imwrite(output_normal, img));
cv::Mat img_jpg_normal = cv::imread(output_normal);
EXPECT_EQ(0, cvtest::norm(img_jpg_rst, img_jpg_normal, NORM_INF));
remove(output_rst.c_str());
}
#endif

6
modules/imgproc/src/contours.cpp
View File

@ -1704,8 +1704,10 @@ void cv::findContours( InputOutputArray _image, OutputArrayOfArrays _contours,
OutputArray _hierarchy, int mode, int method, Point offset )
{
// Sanity check: output must be of type vector<vector<Point>>
CV_Assert( _contours.kind() == _InputArray::STD_VECTOR_VECTOR &&
_contours.channels() == 2 && _contours.depth() == CV_32S );
CV_Assert((_contours.kind() == _InputArray::STD_VECTOR_VECTOR || _contours.kind() == _InputArray::STD_VECTOR_MAT ||
_contours.kind() == _InputArray::STD_VECTOR_UMAT));
CV_Assert(_contours.empty() || (_contours.channels() == 2 && _contours.depth() == CV_32S));
Mat image = _image.getMat();
MemStorage storage(cvCreateMemStorage());

13
modules/imgproc/src/opencl/warp_affine.cl
View File

@ -98,15 +98,15 @@ __kernel void warpAffine(__global const uchar * srcptr, int src_step, int src_of
{
int round_delta = (AB_SCALE >> 1);
int X0_ = rint(M[0] * dx * AB_SCALE);
int Y0_ = rint(M[3] * dx * AB_SCALE);
int X0 = rint(fma(M[0], dx, fma(M[1], dy0, M[2])) * AB_SCALE) + round_delta;
int Y0 = rint(fma(M[3], dx, fma(M[4], dy0, M[5])) * AB_SCALE) + round_delta;
int XSTEP = (int)(M[1] * AB_SCALE);
int YSTEP = (int)(M[4] * AB_SCALE);
int dst_index = mad24(dy0, dst_step, mad24(dx, pixsize, dst_offset));
for (int dy = dy0, dy1 = min(dst_rows, dy0 + rowsPerWI); dy < dy1; ++dy, dst_index += dst_step)
{
int X0 = X0_ + rint(fma(M[1], dy, M[2]) * AB_SCALE) + round_delta;
int Y0 = Y0_ + rint(fma(M[4], dy, M[5]) * AB_SCALE) + round_delta;
short sx = convert_short_sat(X0 >> AB_BITS);
short sy = convert_short_sat(Y0 >> AB_BITS);
@ -117,6 +117,9 @@ __kernel void warpAffine(__global const uchar * srcptr, int src_step, int src_of
}
else
storepix(scalar, dstptr + dst_index);
X0 += XSTEP;
Y0 += YSTEP;
}
}
}

211
modules/objdetect/doc/erfilter.rst
View File

@ -1,211 +0,0 @@
Scene Text Detection
====================
.. highlight:: cpp
Class-specific Extremal Regions for Scene Text Detection
--------------------------------------------------------
The scene text detection algorithm described below has been initially proposed by Lukás Neumann & Jiri Matas [Neumann12]. The main idea behind Class-specific Extremal Regions is similar to the MSER in that suitable Extremal Regions (ERs) are selected from the whole component tree of the image. However, this technique differs from MSER in that selection of suitable ERs is done by a sequential classifier trained for character detection, i.e. dropping the stability requirement of MSERs and selecting class-specific (not necessarily stable) regions.
The component tree of an image is constructed by thresholding by an increasing value step-by-step from 0 to 255 and then linking the obtained connected components from successive levels in a hierarchy by their inclusion relation:
.. image:: pics/component_tree.png
:width: 100%
The component tree may conatain a huge number of regions even for a very simple image as shown in the previous image. This number can easily reach the order of 1 x 10^6 regions for an average 1 Megapixel image. In order to efficiently select suitable regions among all the ERs the algorithm make use of a sequential classifier with two differentiated stages.
In the first stage incrementally computable descriptors (area, perimeter, bounding box, and euler number) are computed (in O(1)) for each region r and used as features for a classifier which estimates the class-conditional probability p(r|character). Only the ERs which correspond to local maximum of the probability p(r|character) are selected (if their probability is above a global limit p_min and the difference between local maximum and local minimum is greater than a \delta_min value).
In the second stage, the ERs that passed the first stage are classified into character and non-character classes using more informative but also more computationally expensive features. (Hole area ratio, convex hull ratio, and the number of outer boundary inflexion points).
This ER filtering process is done in different single-channel projections of the input image in order to increase the character localization recall.
After the ER filtering is done on each input channel, character candidates must be grouped in high-level text blocks (i.e. words, text lines, paragraphs, ...). The grouping algorithm used in this implementation has been proposed by Lluis Gomez and Dimosthenis Karatzas in [Gomez13] and basically consist in finding meaningful groups of regions using a perceptual organization based clustering analisys (see :ocv:func:`erGrouping`).
To see the text detector at work, have a look at the textdetection demo: https://github.com/Itseez/opencv/blob/master/samples/cpp/textdetection.cpp
.. [Neumann12] Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012. The paper is available online at http://cmp.felk.cvut.cz/~neumalu1/neumann-cvpr2012.pdf
.. [Gomez13] Gomez L. and Karatzas D.: Multi-script Text Extraction from Natural Scenes, ICDAR 2013. The paper is available online at http://158.109.8.37/files/GoK2013.pdf
ERStat
------
.. ocv:struct:: ERStat
The ERStat structure represents a class-specific Extremal Region (ER).
An ER is a 4-connected set of pixels with all its grey-level values smaller than the values in its outer boundary. A class-specific ER is selected (using a classifier) from all the ER's in the component tree of the image. ::
struct CV_EXPORTS ERStat
{
public:
//! Constructor
explicit ERStat(int level = 256, int pixel = 0, int x = 0, int y = 0);
//! Destructor
~ERStat() { }
//! seed point and threshold (max grey-level value)
int pixel;
int level;
//! incrementally computable features
int area;
int perimeter;
int euler; //!< euler number
Rect rect; //!< bounding box
double raw_moments[2]; //!< order 1 raw moments to derive the centroid
double central_moments[3]; //!< order 2 central moments to construct the covariance matrix
std::deque<int> *crossings;//!< horizontal crossings
float med_crossings; //!< median of the crossings at three different height levels
//! 2nd stage features
float hole_area_ratio;
float convex_hull_ratio;
float num_inflexion_points;
//! probability that the ER belongs to the class we are looking for
double probability;
//! pointers preserving the tree structure of the component tree
ERStat* parent;
ERStat* child;
ERStat* next;
ERStat* prev;
};
computeNMChannels
-----------------
Compute the different channels to be processed independently in the N&M algorithm [Neumann12].
.. ocv:function:: void computeNMChannels(InputArray _src, OutputArrayOfArrays _channels, int _mode = ERFILTER_NM_RGBLGrad)
:param _src: Source image. Must be RGB ``CV_8UC3``.
:param _channels: Output vector<Mat> where computed channels are stored.
:param _mode: Mode of operation. Currently the only available options are: **ERFILTER_NM_RGBLGrad** (used by default) and **ERFILTER_NM_IHSGrad**.
In N&M algorithm, the combination of intensity (I), hue (H), saturation (S), and gradient magnitude channels (Grad) are used in order to obtain high localization recall. This implementation also provides an alternative combination of red (R), green (G), blue (B), lightness (L), and gradient magnitude (Grad).
ERFilter
--------
.. ocv:class:: ERFilter : public Algorithm
Base class for 1st and 2nd stages of Neumann and Matas scene text detection algorithm [Neumann12]. ::
class CV_EXPORTS ERFilter : public Algorithm
{
public:
//! callback with the classifier is made a class.
//! By doing it we hide SVM, Boost etc. Developers can provide their own classifiers
class CV_EXPORTS Callback
{
public:
virtual ~Callback() { }
//! The classifier must return probability measure for the region.
virtual double eval(const ERStat& stat) = 0;
};
/*!
the key method. Takes image on input and returns the selected regions in a vector of ERStat
only distinctive ERs which correspond to characters are selected by a sequential classifier
*/
virtual void run( InputArray image, std::vector<ERStat>& regions ) = 0;
(...)
};
ERFilter::Callback
------------------
Callback with the classifier is made a class. By doing it we hide SVM, Boost etc. Developers can provide their own classifiers to the ERFilter algorithm.
.. ocv:class:: ERFilter::Callback
ERFilter::Callback::eval
------------------------
The classifier must return probability measure for the region.
.. ocv:function:: double ERFilter::Callback::eval(const ERStat& stat)
:param stat: The region to be classified
ERFilter::run
-------------
The key method of ERFilter algorithm. Takes image on input and returns the selected regions in a vector of ERStat only distinctive ERs which correspond to characters are selected by a sequential classifier
.. ocv:function:: void ERFilter::run( InputArray image, std::vector<ERStat>& regions )
:param image: Sinle channel image ``CV_8UC1``
:param regions: Output for the 1st stage and Input/Output for the 2nd. The selected Extremal Regions are stored here.
Extracts the component tree (if needed) and filter the extremal regions (ER's) by using a given classifier.
createERFilterNM1
-----------------
Create an Extremal Region Filter for the 1st stage classifier of N&M algorithm [Neumann12].
.. ocv:function:: Ptr<ERFilter> createERFilterNM1( const Ptr<ERFilter::Callback>& cb, int thresholdDelta = 1, float minArea = 0.00025, float maxArea = 0.13, float minProbability = 0.4, bool nonMaxSuppression = true, float minProbabilityDiff = 0.1 )
:param cb: Callback with the classifier. Default classifier can be implicitly load with function :ocv:func:`loadClassifierNM1`, e.g. from file in samples/cpp/trained_classifierNM1.xml
:param thresholdDelta: Threshold step in subsequent thresholds when extracting the component tree
:param minArea: The minimum area (% of image size) allowed for retreived ER's
:param minArea: The maximum area (% of image size) allowed for retreived ER's
:param minProbability: The minimum probability P(er|character) allowed for retreived ER's
:param nonMaxSuppression: Whenever non-maximum suppression is done over the branch probabilities
:param minProbability: The minimum probability difference between local maxima and local minima ERs
The component tree of the image is extracted by a threshold increased step by step from 0 to 255, incrementally computable descriptors (aspect_ratio, compactness, number of holes, and number of horizontal crossings) are computed for each ER and used as features for a classifier which estimates the class-conditional probability P(er|character). The value of P(er|character) is tracked using the inclusion relation of ER across all thresholds and only the ERs which correspond to local maximum of the probability P(er|character) are selected (if the local maximum of the probability is above a global limit pmin and the difference between local maximum and local minimum is greater than minProbabilityDiff).
createERFilterNM2
-----------------
Create an Extremal Region Filter for the 2nd stage classifier of N&M algorithm [Neumann12].
.. ocv:function:: Ptr<ERFilter> createERFilterNM2( const Ptr<ERFilter::Callback>& cb, float minProbability = 0.3 )
:param cb: Callback with the classifier. Default classifier can be implicitly load with function :ocv:func:`loadClassifierNM2`, e.g. from file in samples/cpp/trained_classifierNM2.xml
:param minProbability: The minimum probability P(er|character) allowed for retreived ER's
In the second stage, the ERs that passed the first stage are classified into character and non-character classes using more informative but also more computationally expensive features. The classifier uses all the features calculated in the first stage and the following additional features: hole area ratio, convex hull ratio, and number of outer inflexion points.
loadClassifierNM1
-----------------
Allow to implicitly load the default classifier when creating an ERFilter object.
.. ocv:function:: Ptr<ERFilter::Callback> loadClassifierNM1(const std::string& filename)
:param filename: The XML or YAML file with the classifier model (e.g. trained_classifierNM1.xml)
returns a pointer to ERFilter::Callback.
loadClassifierNM2
-----------------
Allow to implicitly load the default classifier when creating an ERFilter object.
.. ocv:function:: Ptr<ERFilter::Callback> loadClassifierNM2(const std::string& filename)
:param filename: The XML or YAML file with the classifier model (e.g. trained_classifierNM2.xml)
returns a pointer to ERFilter::Callback.
erGrouping
----------
Find groups of Extremal Regions that are organized as text blocks.
.. ocv:function:: void erGrouping( InputArrayOfArrays src, std::vector<std::vector<ERStat> > &regions, const std::string& filename, float minProbablity, std::vector<Rect > &groups)
:param src: Vector of sinle channel images CV_8UC1 from wich the regions were extracted
:param regions: Vector of ER's retreived from the ERFilter algorithm from each channel
:param filename: The XML or YAML file with the classifier model (e.g. trained_classifier_erGrouping.xml)
:param minProbability: The minimum probability for accepting a group
:param groups: The output of the algorithm are stored in this parameter as list of rectangles.
This function implements the grouping algorithm described in [Gomez13]. Notice that this implementation constrains the results to horizontally-aligned text and latin script (since ERFilter classifiers are trained only for latin script detection).
The algorithm combines two different clustering techniques in a single parameter-free procedure to detect groups of regions organized as text. The maximally meaningful groups are fist detected in several feature spaces, where each feature space is a combination of proximity information (x,y coordinates) and a similarity measure (intensity, color, size, gradient magnitude, etc.), thus providing a set of hypotheses of text groups. Evidence Accumulation framework is used to combine all these hypotheses to get the final estimate. Each of the resulting groups are finally validated using a classifier in order to assess if they form a valid horizontally-aligned text block.

262
modules/objdetect/doc/latent_svm.rst
View File

@ -1,262 +0,0 @@
Latent SVM
===============================================================
Discriminatively Trained Part Based Models for Object Detection
---------------------------------------------------------------
The object detector described below has been initially proposed by
P.F. Felzenszwalb in [Felzenszwalb2010]_. It is based on a
Dalal-Triggs detector that uses a single filter on histogram of
oriented gradients (HOG) features to represent an object category.
This detector uses a sliding window approach, where a filter is
applied at all positions and scales of an image. The first
innovation is enriching the Dalal-Triggs model using a
star-structured part-based model defined by a "root" filter
(analogous to the Dalal-Triggs filter) plus a set of parts filters
and associated deformation models. The score of one of star models
at a particular position and scale within an image is the score of
the root filter at the given location plus the sum over parts of the
maximum, over placements of that part, of the part filter score on
its location minus a deformation cost easuring the deviation of the
part from its ideal location relative to the root. Both root and
part filter scores are defined by the dot product between a filter
(a set of weights) and a subwindow of a feature pyramid computed
from the input image. Another improvement is a representation of the
class of models by a mixture of star models. The score of a mixture
model at a particular position and scale is the maximum over
components, of the score of that component model at the given
location.
In OpenCV there are C implementation of Latent SVM and C++ wrapper of it.
C version is the structure :ocv:struct:`CvObjectDetection` and a set of functions
working with this structure (see :ocv:func:`cvLoadLatentSvmDetector`,
:ocv:func:`cvReleaseLatentSvmDetector`, :ocv:func:`cvLatentSvmDetectObjects`).
C++ version is the class :ocv:class:`LatentSvmDetector` and has slightly different
functionality in contrast with C version - it supports loading and detection
of several models.
There are two examples of Latent SVM usage: ``samples/c/latentsvmdetect.cpp``
and ``samples/cpp/latentsvm_multidetect.cpp``.
.. highlight:: c
CvLSVMFilterPosition
--------------------
.. ocv:struct:: CvLSVMFilterPosition
Structure describes the position of the filter in the feature pyramid.
.. ocv:member:: unsigned int l
level in the feature pyramid
.. ocv:member:: unsigned int x
x-coordinate in level l
.. ocv:member:: unsigned int y
y-coordinate in level l
CvLSVMFilterObject
------------------
.. ocv:struct:: CvLSVMFilterObject
Description of the filter, which corresponds to the part of the object.
.. ocv:member:: CvLSVMFilterPosition V
ideal (penalty = 0) position of the partial filter
from the root filter position (V_i in the paper)
.. ocv:member:: float fineFunction[4]
vector describes penalty function (d_i in the paper)
pf[0] * x + pf[1] * y + pf[2] * x^2 + pf[3] * y^2
.. ocv:member:: int sizeX
.. ocv:member:: int sizeY
Rectangular map (sizeX x sizeY),
every cell stores feature vector (dimension = p)
.. ocv:member:: int numFeatures
number of features
.. ocv:member:: float *H
matrix of feature vectors to set and get
feature vectors (i,j) used formula H[(j * sizeX + i) * p + k],
where k - component of feature vector in cell (i, j)
CvLatentSvmDetector
-------------------
.. ocv:struct:: CvLatentSvmDetector
Structure contains internal representation of trained Latent SVM detector.
.. ocv:member:: int num_filters
total number of filters (root plus part) in model
.. ocv:member:: int num_components
number of components in model
.. ocv:member:: int* num_part_filters
array containing number of part filters for each component
.. ocv:member:: CvLSVMFilterObject** filters
root and part filters for all model components
.. ocv:member:: float* b
biases for all model components
.. ocv:member:: float score_threshold
confidence level threshold
CvObjectDetection
-----------------
.. ocv:struct:: CvObjectDetection
Structure contains the bounding box and confidence level for detected object.
.. ocv:member:: CvRect rect
bounding box for a detected object
.. ocv:member:: float score
confidence level
cvLoadLatentSvmDetector
-----------------------
Loads trained detector from a file.
.. ocv:function:: CvLatentSvmDetector* cvLoadLatentSvmDetector(const char* filename)
:param filename: Name of the file containing the description of a trained detector
cvReleaseLatentSvmDetector
--------------------------
Release memory allocated for CvLatentSvmDetector structure.
.. ocv:function:: void cvReleaseLatentSvmDetector(CvLatentSvmDetector** detector)
:param detector: CvLatentSvmDetector structure to be released
cvLatentSvmDetectObjects
------------------------
Find rectangular regions in the given image that are likely to contain objects
and corresponding confidence levels.
.. ocv:function:: CvSeq* cvLatentSvmDetectObjects( IplImage* image, CvLatentSvmDetector* detector, CvMemStorage* storage, float overlap_threshold=0.5f, int numThreads=-1 )
:param image: image
:param detector: LatentSVM detector in internal representation
:param storage: Memory storage to store the resultant sequence of the object candidate rectangles
:param overlap_threshold: Threshold for the non-maximum suppression algorithm
:param numThreads: Number of threads used in parallel version of the algorithm
.. highlight:: cpp
LatentSvmDetector
-----------------
.. ocv:class:: LatentSvmDetector
This is a C++ wrapping class of Latent SVM. It contains internal representation of several
trained Latent SVM detectors (models) and a set of methods to load the detectors and detect objects
using them.
LatentSvmDetector::ObjectDetection
----------------------------------
.. ocv:struct:: LatentSvmDetector::ObjectDetection
Structure contains the detection information.
.. ocv:member:: Rect rect
bounding box for a detected object
.. ocv:member:: float score
confidence level
.. ocv:member:: int classID
class (model or detector) ID that detect an object
LatentSvmDetector::LatentSvmDetector
------------------------------------
Two types of constructors.
.. ocv:function:: LatentSvmDetector::LatentSvmDetector()
.. ocv:function:: LatentSvmDetector::LatentSvmDetector(const vector<String>& filenames, const vector<String>& classNames=vector<String>())
:param filenames: A set of filenames storing the trained detectors (models). Each file contains one model. See examples of such files here /opencv_extra/testdata/cv/latentsvmdetector/models_VOC2007/.
:param classNames: A set of trained models names. If it's empty then the name of each model will be constructed from the name of file containing the model. E.g. the model stored in "/home/user/cat.xml" will get the name "cat".
LatentSvmDetector::~LatentSvmDetector
-------------------------------------
Destructor.
.. ocv:function:: LatentSvmDetector::~LatentSvmDetector()
LatentSvmDetector::~clear
-------------------------
Clear all trained models and their names stored in an class object.
.. ocv:function:: void LatentSvmDetector::clear()
LatentSvmDetector::load
-----------------------
Load the trained models from given ``.xml`` files and return ``true`` if at least one model was loaded.
.. ocv:function:: bool LatentSvmDetector::load( const vector<String>& filenames, const vector<String>& classNames=vector<String>() )
:param filenames: A set of filenames storing the trained detectors (models). Each file contains one model. See examples of such files here /opencv_extra/testdata/cv/latentsvmdetector/models_VOC2007/.
:param classNames: A set of trained models names. If it's empty then the name of each model will be constructed from the name of file containing the model. E.g. the model stored in "/home/user/cat.xml" will get the name "cat".
LatentSvmDetector::detect
-------------------------
Find rectangular regions in the given image that are likely to contain objects of loaded classes (models)
and corresponding confidence levels.
.. ocv:function:: void LatentSvmDetector::detect( const Mat& image, vector<ObjectDetection>& objectDetections, float overlapThreshold=0.5f, int numThreads=-1 )
:param image: An image.
:param objectDetections: The detections: rectangulars, scores and class IDs.
:param overlapThreshold: Threshold for the non-maximum suppression algorithm.
:param numThreads: Number of threads used in parallel version of the algorithm.
LatentSvmDetector::getClassNames
--------------------------------
Return the class (model) names that were passed in constructor or method ``load`` or extracted from models filenames in those methods.
.. ocv:function:: const vector<String>& LatentSvmDetector::getClassNames() const
LatentSvmDetector::getClassCount
--------------------------------
Return a count of loaded models (classes).
.. ocv:function:: size_t LatentSvmDetector::getClassCount() const
.. [Felzenszwalb2010] Felzenszwalb, P. F. and Girshick, R. B. and McAllester, D. and Ramanan, D. *Object Detection with Discriminatively Trained Part Based Models*. PAMI, vol. 32, no. 9, pp. 1627-1645, September 2010

2
modules/objdetect/doc/objdetect.rst
View File

@ -8,5 +8,3 @@ objdetect. Object Detection
:maxdepth: 2
cascade_classification
latent_svm
erfilter

BIN
modules/objdetect/doc/pics/component_tree.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 106 KiB

2
modules/objdetect/include/opencv2/objdetect.hpp
View File

@ -315,8 +315,6 @@ public:
}
#include "opencv2/objdetect/linemod.hpp"
#include "opencv2/objdetect/erfilter.hpp"
#include "opencv2/objdetect/detection_based_tracker.hpp"
#endif

266
modules/objdetect/include/opencv2/objdetect/erfilter.hpp
View File

@ -1,266 +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.
// Copyright (C) 2013, OpenCV Foundation, 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_OBJDETECT_ERFILTER_HPP__
#define __OPENCV_OBJDETECT_ERFILTER_HPP__
#include "opencv2/core.hpp"
#include <vector>
#include <deque>
#include <string>
namespace cv
{
/*!
Extremal Region Stat structure
The ERStat structure represents a class-specific Extremal Region (ER).
An ER is a 4-connected set of pixels with all its grey-level values smaller than the values
in its outer boundary. A class-specific ER is selected (using a classifier) from all the ER's
in the component tree of the image.
*/
struct CV_EXPORTS ERStat
{
public:
//! Constructor
explicit ERStat(int level = 256, int pixel = 0, int x = 0, int y = 0);
//! Destructor
~ERStat() { }
//! seed point and the threshold (max grey-level value)
int pixel;
int level;
//! incrementally computable features
int area;
int perimeter;
int euler; //!< euler number
Rect rect;
double raw_moments[2]; //!< order 1 raw moments to derive the centroid
double central_moments[3]; //!< order 2 central moments to construct the covariance matrix
std::deque<int> *crossings;//!< horizontal crossings
float med_crossings; //!< median of the crossings at three different height levels
//! 2nd stage features
float hole_area_ratio;
float convex_hull_ratio;
float num_inflexion_points;
// TODO Other features can be added (average color, standard deviation, and such)
// TODO shall we include the pixel list whenever available (i.e. after 2nd stage) ?
std::vector<int> *pixels;
//! probability that the ER belongs to the class we are looking for
double probability;
//! pointers preserving the tree structure of the component tree
ERStat* parent;
ERStat* child;
ERStat* next;
ERStat* prev;
//! wenever the regions is a local maxima of the probability
bool local_maxima;
ERStat* max_probability_ancestor;
ERStat* min_probability_ancestor;
};
/*!
Base class for 1st and 2nd stages of Neumann and Matas scene text detection algorithms
Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012
Extracts the component tree (if needed) and filter the extremal regions (ER's) by using a given classifier.
*/
class CV_EXPORTS ERFilter : public Algorithm
{
public:
//! callback with the classifier is made a class. By doing it we hide SVM, Boost etc.
class CV_EXPORTS Callback
{
public:
virtual ~Callback() { }
//! The classifier must return probability measure for the region.
virtual double eval(const ERStat& stat) = 0; //const = 0; //TODO why cannot use const = 0 here?
};
/*!
the key method. Takes image on input and returns the selected regions in a vector of ERStat
only distinctive ERs which correspond to characters are selected by a sequential classifier
\param image is the input image
\param regions is output for the first stage, input/output for the second one.
*/
virtual void run( InputArray image, std::vector<ERStat>& regions ) = 0;
//! set/get methods to set the algorithm properties,
virtual void setCallback(const Ptr<ERFilter::Callback>& cb) = 0;
virtual void setThresholdDelta(int thresholdDelta) = 0;
virtual void setMinArea(float minArea) = 0;
virtual void setMaxArea(float maxArea) = 0;
virtual void setMinProbability(float minProbability) = 0;
virtual void setMinProbabilityDiff(float minProbabilityDiff) = 0;
virtual void setNonMaxSuppression(bool nonMaxSuppression) = 0;
virtual int getNumRejected() = 0;
};
/*!
Create an Extremal Region Filter for the 1st stage classifier of N&M algorithm
Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012
The component tree of the image is extracted by a threshold increased step by step
from 0 to 255, incrementally computable descriptors (aspect_ratio, compactness,
number of holes, and number of horizontal crossings) are computed for each ER
and used as features for a classifier which estimates the class-conditional
probability P(er|character). The value of P(er|character) is tracked using the inclusion
relation of ER across all thresholds and only the ERs which correspond to local maximum
of the probability P(er|character) are selected (if the local maximum of the
probability is above a global limit pmin and the difference between local maximum and
local minimum is greater than minProbabilityDiff).
\param cb Callback with the classifier.
default classifier can be implicitly load with function loadClassifierNM1()
from file in samples/cpp/trained_classifierNM1.xml
\param thresholdDelta Threshold step in subsequent thresholds when extracting the component tree
\param minArea The minimum area (% of image size) allowed for retreived ER's
\param minArea The maximum area (% of image size) allowed for retreived ER's
\param minProbability The minimum probability P(er|character) allowed for retreived ER's
\param nonMaxSuppression Whenever non-maximum suppression is done over the branch probabilities
\param minProbability The minimum probability difference between local maxima and local minima ERs
*/
CV_EXPORTS Ptr<ERFilter> createERFilterNM1(const Ptr<ERFilter::Callback>& cb,
int thresholdDelta = 1, float minArea = 0.00025,
float maxArea = 0.13, float minProbability = 0.4,
bool nonMaxSuppression = true,
float minProbabilityDiff = 0.1);
/*!
Create an Extremal Region Filter for the 2nd stage classifier of N&M algorithm
Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012
In the second stage, the ERs that passed the first stage are classified into character
and non-character classes using more informative but also more computationally expensive
features. The classifier uses all the features calculated in the first stage and the following
additional features: hole area ratio, convex hull ratio, and number of outer inflexion points.
\param cb Callback with the classifier
default classifier can be implicitly load with function loadClassifierNM2()
from file in samples/cpp/trained_classifierNM2.xml
\param minProbability The minimum probability P(er|character) allowed for retreived ER's
*/
CV_EXPORTS Ptr<ERFilter> createERFilterNM2(const Ptr<ERFilter::Callback>& cb,
float minProbability = 0.3);
/*!
Allow to implicitly load the default classifier when creating an ERFilter object.
The function takes as parameter the XML or YAML file with the classifier model
(e.g. trained_classifierNM1.xml) returns a pointer to ERFilter::Callback.
*/
CV_EXPORTS Ptr<ERFilter::Callback> loadClassifierNM1(const std::string& filename);
/*!
Allow to implicitly load the default classifier when creating an ERFilter object.
The function takes as parameter the XML or YAML file with the classifier model
(e.g. trained_classifierNM1.xml) returns a pointer to ERFilter::Callback.
*/
CV_EXPORTS Ptr<ERFilter::Callback> loadClassifierNM2(const std::string& filename);
// computeNMChannels operation modes
enum { ERFILTER_NM_RGBLGrad = 0,
ERFILTER_NM_IHSGrad = 1
};
/*!
Compute the different channels to be processed independently in the N&M algorithm
Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012
In N&M algorithm, the combination of intensity (I), hue (H), saturation (S), and gradient
magnitude channels (Grad) are used in order to obtain high localization recall.
This implementation also provides an alternative combination of red (R), green (G), blue (B),
lightness (L), and gradient magnitude (Grad).
\param _src Source image. Must be RGB CV_8UC3.
\param _channels Output vector<Mat> where computed channels are stored.
\param _mode Mode of operation. Currently the only available options are
ERFILTER_NM_RGBLGrad (by default) and ERFILTER_NM_IHSGrad.
*/
CV_EXPORTS void computeNMChannels(InputArray _src, OutputArrayOfArrays _channels, int _mode = ERFILTER_NM_RGBLGrad);
/*!
Find groups of Extremal Regions that are organized as text blocks. This function implements
the grouping algorithm described in:
Gomez L. and Karatzas D.: Multi-script Text Extraction from Natural Scenes, ICDAR 2013.
Notice that this implementation constrains the results to horizontally-aligned text and
latin script (since ERFilter classifiers are trained only for latin script detection).
The algorithm combines two different clustering techniques in a single parameter-free procedure
to detect groups of regions organized as text. The maximally meaningful groups are fist detected
in several feature spaces, where each feature space is a combination of proximity information
(x,y coordinates) and a similarity measure (intensity, color, size, gradient magnitude, etc.),
thus providing a set of hypotheses of text groups. Evidence Accumulation framework is used to
combine all these hypotheses to get the final estimate. Each of the resulting groups are finally
validated using a classifier in order to assest if they form a valid horizontally-aligned text block.
\param src Vector of sinle channel images CV_8UC1 from wich the regions were extracted.
\param regions Vector of ER's retreived from the ERFilter algorithm from each channel
\param filename The XML or YAML file with the classifier model (e.g. trained_classifier_erGrouping.xml)
\param minProbability The minimum probability for accepting a group
\param groups The output of the algorithm are stored in this parameter as list of rectangles.
*/
CV_EXPORTS void erGrouping(InputArrayOfArrays src, std::vector<std::vector<ERStat> > &regions,
const std::string& filename, float minProbablity,
std::vector<Rect > &groups);
}
#endif // _OPENCV_ERFILTER_HPP_

455
modules/objdetect/include/opencv2/objdetect/linemod.hpp
View File

@ -1,455 +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.
// Copyright (C) 2013, OpenCV Foundation, 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_OBJDETECT_LINEMOD_HPP__
#define __OPENCV_OBJDETECT_LINEMOD_HPP__
#include "opencv2/core.hpp"
#include <map>
/****************************************************************************************\
* LINE-MOD *
\****************************************************************************************/
namespace cv {
namespace linemod {
/// @todo Convert doxy comments to rst
/**
* \brief Discriminant feature described by its location and label.
*/
struct CV_EXPORTS Feature
{
int x; ///< x offset
int y; ///< y offset
int label; ///< Quantization
Feature() : x(0), y(0), label(0) {}
Feature(int x, int y, int label);
void read(const FileNode& fn);
void write(FileStorage& fs) const;
};
inline Feature::Feature(int _x, int _y, int _label) : x(_x), y(_y), label(_label) {}
struct CV_EXPORTS Template
{
int width;
int height;
int pyramid_level;
std::vector<Feature> features;
void read(const FileNode& fn);
void write(FileStorage& fs) const;
};
/**
* \brief Represents a modality operating over an image pyramid.
*/
class QuantizedPyramid
{
public:
// Virtual destructor
virtual ~QuantizedPyramid() {}
/**
* \brief Compute quantized image at current pyramid level for online detection.
*
* \param[out] dst The destination 8-bit image. For each pixel at most one bit is set,
* representing its classification.
*/
virtual void quantize(Mat& dst) const =0;
/**
* \brief Extract most discriminant features at current pyramid level to form a new template.
*
* \param[out] templ The new template.
*/
virtual bool extractTemplate(Template& templ) const =0;
/**
* \brief Go to the next pyramid level.
*
* \todo Allow pyramid scale factor other than 2
*/
virtual void pyrDown() =0;
protected:
/// Candidate feature with a score
struct Candidate
{
Candidate(int x, int y, int label, float score);
/// Sort candidates with high score to the front
bool operator<(const Candidate& rhs) const
{
return score > rhs.score;
}
Feature f;
float score;
};
/**
* \brief Choose candidate features so that they are not bunched together.
*
* \param[in] candidates Candidate features sorted by score.
* \param[out] features Destination vector of selected features.
* \param[in] num_features Number of candidates to select.
* \param[in] distance Hint for desired distance between features.
*/
static void selectScatteredFeatures(const std::vector<Candidate>& candidates,
std::vector<Feature>& features,
size_t num_features, float distance);
};
inline QuantizedPyramid::Candidate::Candidate(int x, int y, int label, float _score) : f(x, y, label), score(_score) {}
/**
* \brief Interface for modalities that plug into the LINE template matching representation.
*
* \todo Max response, to allow optimization of summing (255/MAX) features as uint8
*/
class CV_EXPORTS Modality
{
public:
// Virtual destructor
virtual ~Modality() {}
/**
* \brief Form a quantized image pyramid from a source image.
*
* \param[in] src The source image. Type depends on the modality.
* \param[in] mask Optional mask. If not empty, unmasked pixels are set to zero
* in quantized image and cannot be extracted as features.
*/
Ptr<QuantizedPyramid> process(const Mat& src,
const Mat& mask = Mat()) const
{
return processImpl(src, mask);
}
virtual String name() const =0;
virtual void read(const FileNode& fn) =0;
virtual void write(FileStorage& fs) const =0;
/**
* \brief Create modality by name.
*
* The following modality types are supported:
* - "ColorGradient"
* - "DepthNormal"
*/
static Ptr<Modality> create(const String& modality_type);
/**
* \brief Load a modality from file.
*/
static Ptr<Modality> create(const FileNode& fn);
protected:
// Indirection is because process() has a default parameter.
virtual Ptr<QuantizedPyramid> processImpl(const Mat& src,
const Mat& mask) const =0;
};
/**
* \brief Modality that computes quantized gradient orientations from a color image.
*/
class CV_EXPORTS ColorGradient : public Modality
{
public:
/**
* \brief Default constructor. Uses reasonable default parameter values.
*/
ColorGradient();
/**
* \brief Constructor.
*
* \param weak_threshold When quantizing, discard gradients with magnitude less than this.
* \param num_features How many features a template must contain.
* \param strong_threshold Consider as candidate features only gradients whose norms are
* larger than this.
*/
ColorGradient(float weak_threshold, size_t num_features, float strong_threshold);
virtual String name() const;
virtual void read(const FileNode& fn);
virtual void write(FileStorage& fs) const;
float weak_threshold;
size_t num_features;
float strong_threshold;
protected:
virtual Ptr<QuantizedPyramid> processImpl(const Mat& src,
const Mat& mask) const;
};
/**
* \brief Modality that computes quantized surface normals from a dense depth map.
*/
class CV_EXPORTS DepthNormal : public Modality
{
public:
/**
* \brief Default constructor. Uses reasonable default parameter values.
*/
DepthNormal();
/**
* \brief Constructor.
*
* \param distance_threshold Ignore pixels beyond this distance.
* \param difference_threshold When computing normals, ignore contributions of pixels whose
* depth difference with the central pixel is above this threshold.
* \param num_features How many features a template must contain.
* \param extract_threshold Consider as candidate feature only if there are no differing
* orientations within a distance of extract_threshold.
*/
DepthNormal(int distance_threshold, int difference_threshold, size_t num_features,
int extract_threshold);
virtual String name() const;
virtual void read(const FileNode& fn);
virtual void write(FileStorage& fs) const;
int distance_threshold;
int difference_threshold;
size_t num_features;
int extract_threshold;
protected:
virtual Ptr<QuantizedPyramid> processImpl(const Mat& src,
const Mat& mask) const;
};
/**
* \brief Debug function to colormap a quantized image for viewing.
*/
void colormap(const Mat& quantized, Mat& dst);
/**
* \brief Represents a successful template match.
*/
struct CV_EXPORTS Match
{
Match()
{
}
Match(int x, int y, float similarity, const String& class_id, int template_id);
/// Sort matches with high similarity to the front
bool operator<(const Match& rhs) const
{
// Secondarily sort on template_id for the sake of duplicate removal
if (similarity != rhs.similarity)
return similarity > rhs.similarity;
else
return template_id < rhs.template_id;
}
bool operator==(const Match& rhs) const
{
return x == rhs.x && y == rhs.y && similarity == rhs.similarity && class_id == rhs.class_id;
}
int x;
int y;
float similarity;
String class_id;
int template_id;
};
inline
Match::Match(int _x, int _y, float _similarity, const String& _class_id, int _template_id)
: x(_x), y(_y), similarity(_similarity), class_id(_class_id), template_id(_template_id)
{}
/**
* \brief Object detector using the LINE template matching algorithm with any set of
* modalities.
*/
class CV_EXPORTS Detector
{
public:
/**
* \brief Empty constructor, initialize with read().
*/
Detector();
/**
* \brief Constructor.
*
* \param modalities Modalities to use (color gradients, depth normals, ...).
* \param T_pyramid Value of the sampling step T at each pyramid level. The
* number of pyramid levels is T_pyramid.size().
*/
Detector(const std::vector< Ptr<Modality> >& modalities, const std::vector<int>& T_pyramid);
/**
* \brief Detect objects by template matching.
*
* Matches globally at the lowest pyramid level, then refines locally stepping up the pyramid.
*
* \param sources Source images, one for each modality.
* \param threshold Similarity threshold, a percentage between 0 and 100.
* \param[out] matches Template matches, sorted by similarity score.
* \param class_ids If non-empty, only search for the desired object classes.
* \param[out] quantized_images Optionally return vector<Mat> of quantized images.
* \param masks The masks for consideration during matching. The masks should be CV_8UC1
* where 255 represents a valid pixel. If non-empty, the vector must be
* the same size as sources. Each element must be
* empty or the same size as its corresponding source.
*/
void match(const std::vector<Mat>& sources, float threshold, std::vector<Match>& matches,
const std::vector<String>& class_ids = std::vector<String>(),
OutputArrayOfArrays quantized_images = noArray(),
const std::vector<Mat>& masks = std::vector<Mat>()) const;
/**
* \brief Add new object template.
*
* \param sources Source images, one for each modality.
* \param class_id Object class ID.
* \param object_mask Mask separating object from background.
* \param[out] bounding_box Optionally return bounding box of the extracted features.
*
* \return Template ID, or -1 if failed to extract a valid template.
*/
int addTemplate(const std::vector<Mat>& sources, const String& class_id,
const Mat& object_mask, Rect* bounding_box = NULL);
/**
* \brief Add a new object template computed by external means.
*/
int addSyntheticTemplate(const std::vector<Template>& templates, const String& class_id);
/**
* \brief Get the modalities used by this detector.
*
* You are not permitted to add/remove modalities, but you may dynamic_cast them to
* tweak parameters.
*/
const std::vector< Ptr<Modality> >& getModalities() const { return modalities; }
/**
* \brief Get sampling step T at pyramid_level.
*/
int getT(int pyramid_level) const { return T_at_level[pyramid_level]; }
/**
* \brief Get number of pyramid levels used by this detector.
*/
int pyramidLevels() const { return pyramid_levels; }
/**
* \brief Get the template pyramid identified by template_id.
*
* For example, with 2 modalities (Gradient, Normal) and two pyramid levels
* (L0, L1), the order is (GradientL0, NormalL0, GradientL1, NormalL1).
*/
const std::vector<Template>& getTemplates(const String& class_id, int template_id) const;
int numTemplates() const;
int numTemplates(const String& class_id) const;
int numClasses() const { return static_cast<int>(class_templates.size()); }
std::vector<String> classIds() const;
void read(const FileNode& fn);
void write(FileStorage& fs) const;
String readClass(const FileNode& fn, const String &class_id_override = "");
void writeClass(const String& class_id, FileStorage& fs) const;
void readClasses(const std::vector<String>& class_ids,
const String& format = "templates_%s.yml.gz");
void writeClasses(const String& format = "templates_%s.yml.gz") const;
protected:
std::vector< Ptr<Modality> > modalities;
int pyramid_levels;
std::vector<int> T_at_level;
typedef std::vector<Template> TemplatePyramid;
typedef std::map<String, std::vector<TemplatePyramid> > TemplatesMap;
TemplatesMap class_templates;
typedef std::vector<Mat> LinearMemories;
// Indexed as [pyramid level][modality][quantized label]
typedef std::vector< std::vector<LinearMemories> > LinearMemoryPyramid;
void matchClass(const LinearMemoryPyramid& lm_pyramid,
const std::vector<Size>& sizes,
float threshold, std::vector<Match>& matches,
const String& class_id,
const std::vector<TemplatePyramid>& template_pyramids) const;
};
/**
* \brief Factory function for detector using LINE algorithm with color gradients.
*
* Default parameter settings suitable for VGA images.
*/
CV_EXPORTS Ptr<Detector> getDefaultLINE();
/**
* \brief Factory function for detector using LINE-MOD algorithm with color gradients
* and depth normals.
*
* Default parameter settings suitable for VGA images.
*/
CV_EXPORTS Ptr<Detector> getDefaultLINEMOD();
} // namespace linemod
} // namespace cv
#endif // __OPENCV_OBJDETECT_LINEMOD_HPP__

3187
modules/objdetect/src/erfilter.cpp
View File

File diff suppressed because it is too large Load Diff

1844
modules/objdetect/src/linemod.cpp
View File

File diff suppressed because it is too large Load Diff

4
modules/objdetect/src/normal_lut.i
View File

File diff suppressed because one or more lines are too long

2
modules/photo/src/calibrate.cpp
View File

@ -104,7 +104,7 @@ public:
for(size_t i = 0; i < sample_points.size(); i++) {
for(size_t j = 0; j < images.size(); j++) {
int val = images[j].ptr()[3*(sample_points[i].y * images[j].cols + sample_points[j].x) + channel];
int val = images[j].ptr()[3*(sample_points[i].y * images[j].cols + sample_points[i].x) + channel];
A.at<float>(eq, val) = w.at<float>(val);
A.at<float>(eq, LDR_SIZE + (int)i) = -w.at<float>(val);
B.at<float>(eq, 0) = w.at<float>(val) * log(times.at<float>((int)j));

58
modules/python/CMakeLists.txt
View File

@ -11,7 +11,25 @@ if(ANDROID OR IOS OR NOT PYTHONLIBS_FOUND OR NOT PYTHON_NUMPY_INCLUDE_DIRS)
endif()
set(the_description "The python bindings")
ocv_add_module(python BINDINGS opencv_core opencv_flann opencv_imgproc opencv_video opencv_ml opencv_features2d opencv_imgcodecs opencv_videoio opencv_highgui opencv_calib3d opencv_photo opencv_objdetect OPTIONAL opencv_nonfree)
set(candidate_deps "")
foreach(mp ${OPENCV_MODULES_PATH} ${OPENCV_EXTRA_MODULES_PATH})
file(GLOB names "${mp}/*")
foreach(m IN LISTS names)
if(IS_DIRECTORY ${m})
get_filename_component(m ${m} NAME)
list(APPEND candidate_deps "opencv_${m}")
endif()
endforeach(m)
endforeach(mp)
# module blacklist
ocv_list_filterout(candidate_deps "^opencv_cud(a|ev)")
ocv_list_filterout(candidate_deps "^opencv_adas$")
ocv_list_filterout(candidate_deps "^opencv_tracking$")
ocv_add_module(python BINDINGS OPTIONAL ${candidate_deps})
ocv_module_include_directories(
"${PYTHON_INCLUDE_PATH}"
@ -19,31 +37,17 @@ ocv_module_include_directories(
"${CMAKE_CURRENT_SOURCE_DIR}/src2"
)
set(opencv_hdrs
"${OPENCV_MODULE_opencv_core_LOCATION}/include/opencv2/core.hpp"
"${OPENCV_MODULE_opencv_core_LOCATION}/include/opencv2/core/base.hpp"
"${OPENCV_MODULE_opencv_core_LOCATION}/include/opencv2/core/types.hpp"
"${OPENCV_MODULE_opencv_core_LOCATION}/include/opencv2/core/persistence.hpp"
"${OPENCV_MODULE_opencv_core_LOCATION}/include/opencv2/core/utility.hpp"
"${OPENCV_MODULE_opencv_core_LOCATION}/include/opencv2/core/ocl.hpp"
"${OPENCV_MODULE_opencv_flann_LOCATION}/include/opencv2/flann/miniflann.hpp"
"${OPENCV_MODULE_opencv_imgproc_LOCATION}/include/opencv2/imgproc.hpp"
"${OPENCV_MODULE_opencv_video_LOCATION}/include/opencv2/video/background_segm.hpp"
"${OPENCV_MODULE_opencv_video_LOCATION}/include/opencv2/video/tracking.hpp"
"${OPENCV_MODULE_opencv_photo_LOCATION}/include/opencv2/photo.hpp"
"${OPENCV_MODULE_opencv_imgcodecs_LOCATION}/include/opencv2/imgcodecs.hpp"
"${OPENCV_MODULE_opencv_videoio_LOCATION}/include/opencv2/videoio.hpp"
"${OPENCV_MODULE_opencv_highgui_LOCATION}/include/opencv2/highgui.hpp"
"${OPENCV_MODULE_opencv_ml_LOCATION}/include/opencv2/ml.hpp"
"${OPENCV_MODULE_opencv_features2d_LOCATION}/include/opencv2/features2d.hpp"
"${OPENCV_MODULE_opencv_calib3d_LOCATION}/include/opencv2/calib3d.hpp"
"${OPENCV_MODULE_opencv_objdetect_LOCATION}/include/opencv2/objdetect.hpp"
)
if(HAVE_opencv_nonfree)
list(APPEND opencv_hdrs "${OPENCV_MODULE_opencv_nonfree_LOCATION}/include/opencv2/nonfree/features2d.hpp"
"${OPENCV_MODULE_opencv_nonfree_LOCATION}/include/opencv2/nonfree.hpp")
endif()
set(opencv_hdrs "")
foreach(m IN LISTS OPENCV_MODULE_opencv_python_DEPS)
list(APPEND opencv_hdrs ${OPENCV_MODULE_${m}_HEADERS})
endforeach(m)
# header blacklist
ocv_list_filterout(opencv_hdrs ".h$")
ocv_list_filterout(opencv_hdrs "opencv2/core/cuda")
ocv_list_filterout(opencv_hdrs "opencv2/objdetect/detection_based_tracker.hpp")
ocv_list_filterout(opencv_hdrs "opencv2/optim.hpp")
set(cv2_generated_hdrs
"${CMAKE_CURRENT_BINARY_DIR}/pyopencv_generated_include.h"
@ -53,11 +57,13 @@ set(cv2_generated_hdrs
"${CMAKE_CURRENT_BINARY_DIR}/pyopencv_generated_type_reg.h"
"${CMAKE_CURRENT_BINARY_DIR}/pyopencv_generated_const_reg.h")
file(WRITE "${CMAKE_CURRENT_BINARY_DIR}/headers.txt" "${opencv_hdrs}")
add_custom_command(
OUTPUT ${cv2_generated_hdrs}
COMMAND ${PYTHON_EXECUTABLE} "${CMAKE_CURRENT_SOURCE_DIR}/src2/gen2.py" ${CMAKE_CURRENT_BINARY_DIR} ${opencv_hdrs}
COMMAND ${PYTHON_EXECUTABLE} "${CMAKE_CURRENT_SOURCE_DIR}/src2/gen2.py" ${CMAKE_CURRENT_BINARY_DIR} "${CMAKE_CURRENT_BINARY_DIR}/headers.txt"
DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/src2/gen2.py
DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/src2/hdr_parser.py
DEPENDS ${CMAKE_CURRENT_BINARY_DIR}/headers.txt
DEPENDS ${opencv_hdrs})
add_library(${the_module} SHARED src2/cv2.cpp ${cv2_generated_hdrs})

6
modules/python/src2/gen2.py
View File

@ -831,8 +831,10 @@ class PythonWrapperGenerator(object):
# step 1: scan the headers and build more descriptive maps of classes, consts, functions
for hdr in srcfiles:
self.code_include.write( '#include "{}"\n'.format(hdr[hdr.rindex('opencv2/'):]) )
decls = parser.parse(hdr)
if len(decls) == 0:
continue
self.code_include.write( '#include "{}"\n'.format(hdr[hdr.rindex('opencv2/'):]) )
for decl in decls:
name = decl[0]
if name.startswith("struct") or name.startswith("class"):
@ -901,6 +903,6 @@ if __name__ == "__main__":
if len(sys.argv) > 1:
dstdir = sys.argv[1]
if len(sys.argv) > 2:
srcfiles = sys.argv[2:]
srcfiles = open(sys.argv[2], 'r').read().split(';')
generator = PythonWrapperGenerator()
generator.gen(srcfiles, dstdir)

2
modules/python/src2/hdr_parser.py
View File

@ -206,6 +206,8 @@ class CppHeaderParser(object):
def parse_enum(self, decl_str):
l = decl_str
ll = l.split(",")
if ll[-1].strip() == "":
ll = ll[:-1]
prev_val = ""
prev_val_delta = -1
decl = []

8
modules/shape/include/opencv2/shape/hist_cost.hpp
View File

@ -73,7 +73,7 @@ public:
};
CV_EXPORTS_W Ptr<HistogramCostExtractor>
createNormHistogramCostExtractor(int flag=DIST_L2, int nDummies=25, float defaultCost=0.2);
createNormHistogramCostExtractor(int flag=DIST_L2, int nDummies=25, float defaultCost=0.2f);
/*! */
class CV_EXPORTS_W EMDHistogramCostExtractor : public HistogramCostExtractor
@ -84,20 +84,20 @@ public:
};
CV_EXPORTS_W Ptr<HistogramCostExtractor>
createEMDHistogramCostExtractor(int flag=DIST_L2, int nDummies=25, float defaultCost=0.2);
createEMDHistogramCostExtractor(int flag=DIST_L2, int nDummies=25, float defaultCost=0.2f);
/*! */
class CV_EXPORTS_W ChiHistogramCostExtractor : public HistogramCostExtractor
{};
CV_EXPORTS_W Ptr<HistogramCostExtractor> createChiHistogramCostExtractor(int nDummies=25, float defaultCost=0.2);
CV_EXPORTS_W Ptr<HistogramCostExtractor> createChiHistogramCostExtractor(int nDummies=25, float defaultCost=0.2f);
/*! */
class CV_EXPORTS_W EMDL1HistogramCostExtractor : public HistogramCostExtractor
{};
CV_EXPORTS_W Ptr<HistogramCostExtractor>
createEMDL1HistogramCostExtractor(int nDummies=25, float defaultCost=0.2);
createEMDL1HistogramCostExtractor(int nDummies=25, float defaultCost=0.2f);
} // cv
#endif

4
modules/shape/include/opencv2/shape/shape_distance.hpp
View File

@ -116,7 +116,7 @@ public:
/* Complete constructor */
CV_EXPORTS_W Ptr<ShapeContextDistanceExtractor>
createShapeContextDistanceExtractor(int nAngularBins=12, int nRadialBins=4,
float innerRadius=0.2, float outerRadius=2, int iterations=3,
float innerRadius=0.2f, float outerRadius=2, int iterations=3,
const Ptr<HistogramCostExtractor> &comparer = createChiHistogramCostExtractor(),
const Ptr<ShapeTransformer> &transformer = createThinPlateSplineShapeTransformer());
@ -137,7 +137,7 @@ public:
};
/* Constructor */
CV_EXPORTS_W Ptr<HausdorffDistanceExtractor> createHausdorffDistanceExtractor(int distanceFlag=cv::NORM_L2, float rankProp=0.6);
CV_EXPORTS_W Ptr<HausdorffDistanceExtractor> createHausdorffDistanceExtractor(int distanceFlag=cv::NORM_L2, float rankProp=0.6f);
} // cv
#endif

14
modules/stitching/include/opencv2/stitching/detail/matchers.hpp
View File

@ -183,6 +183,20 @@ protected:
Ptr<FeaturesMatcher> impl_;
};
class CV_EXPORTS BestOf2NearestRangeMatcher : public BestOf2NearestMatcher
{
public:
BestOf2NearestRangeMatcher(int range_width = 5, bool try_use_gpu = false, float match_conf = 0.3f,
int num_matches_thresh1 = 6, int num_matches_thresh2 = 6);
void operator ()(const std::vector<ImageFeatures> &features, std::vector<MatchesInfo> &pairwise_matches,
const cv::UMat &mask = cv::UMat());
protected:
int range_width_;
};
} // namespace detail
} // namespace cv

86
modules/stitching/include/opencv2/stitching/detail/timelapsers.hpp Normal file
View File

@ -0,0 +1,86 @@
/*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_STITCHING_TIMELAPSERS_HPP__
#define __OPENCV_STITCHING_TIMELAPSERS_HPP__
#include "opencv2/core.hpp"
namespace cv {
namespace detail {
// Base Timelapser class, takes a sequence of images, applies appropriate shift, stores result in dst_.
class CV_EXPORTS Timelapser
{
public:
enum {AS_IS, CROP};
virtual ~Timelapser() {}
static Ptr<Timelapser> createDefault(int type);
virtual void initialize(const std::vector<Point> &corners, const std::vector<Size> &sizes);
virtual void process(InputArray img, InputArray mask, Point tl);
virtual const UMat& getDst() {return dst_;}
protected:
virtual bool test_point(Point pt);
UMat dst_;
Rect dst_roi_;
};
class CV_EXPORTS TimelapserCrop : public Timelapser
{
public:
virtual void initialize(const std::vector<Point> &corners, const std::vector<Size> &sizes);
};
} // namespace detail
} // namespace cv
#endif // __OPENCV_STITCHING_TIMELAPSERS_HPP__

3
modules/stitching/include/opencv2/stitching/detail/util.hpp
View File

@ -46,7 +46,9 @@
#include <list>
#include "opencv2/core.hpp"
#ifndef ENABLE_LOG
#define ENABLE_LOG 0
#endif
// TODO remove LOG macros, add logging class
#if ENABLE_LOG
@ -148,6 +150,7 @@ private:
CV_EXPORTS bool overlapRoi(Point tl1, Point tl2, Size sz1, Size sz2, Rect &roi);
CV_EXPORTS Rect resultRoi(const std::vector<Point> &corners, const std::vector<UMat> &images);
CV_EXPORTS Rect resultRoi(const std::vector<Point> &corners, const std::vector<Size> &sizes);
CV_EXPORTS Rect resultRoiIntersection(const std::vector<Point> &corners, const std::vector<Size> &sizes);
CV_EXPORTS Point resultTl(const std::vector<Point> &corners);
// Returns random 'count' element subset of the {0,1,...,size-1} set

34
modules/stitching/src/matchers.cpp
View File

@ -646,5 +646,39 @@ void BestOf2NearestMatcher::collectGarbage()
impl_->collectGarbage();
}
BestOf2NearestRangeMatcher::BestOf2NearestRangeMatcher(int range_width, bool try_use_gpu, float match_conf, int num_matches_thresh1, int num_matches_thresh2): BestOf2NearestMatcher(try_use_gpu, match_conf, num_matches_thresh1, num_matches_thresh2)
{
range_width_ = range_width;
}
void BestOf2NearestRangeMatcher::operator ()(const std::vector<ImageFeatures> &features, std::vector<MatchesInfo> &pairwise_matches,
const UMat &mask)
{
const int num_images = static_cast<int>(features.size());
CV_Assert(mask.empty() || (mask.type() == CV_8U && mask.cols == num_images && mask.rows));
Mat_<uchar> mask_(mask.getMat(ACCESS_READ));
if (mask_.empty())
mask_ = Mat::ones(num_images, num_images, CV_8U);
std::vector<std::pair<int,int> > near_pairs;
for (int i = 0; i < num_images - 1; ++i)
for (int j = i + 1; j < std::min(num_images, i + range_width_); ++j)
if (features[i].keypoints.size() > 0 && features[j].keypoints.size() > 0 && mask_(i, j))
near_pairs.push_back(std::make_pair(i, j));
pairwise_matches.resize(num_images * num_images);
MatchPairsBody body(*this, features, pairwise_matches, near_pairs);
if (is_thread_safe_)
parallel_for_(Range(0, static_cast<int>(near_pairs.size())), body);
else
body(Range(0, static_cast<int>(near_pairs.size())));
LOGLN_CHAT("");
}
} // namespace detail
} // namespace cv

1
modules/stitching/src/precomp.hpp
View File

@ -59,6 +59,7 @@
#include "opencv2/stitching.hpp"
#include "opencv2/stitching/detail/autocalib.hpp"
#include "opencv2/stitching/detail/blenders.hpp"
#include "opencv2/stitching/detail/timelapsers.hpp"
#include "opencv2/stitching/detail/camera.hpp"
#include "opencv2/stitching/detail/exposure_compensate.hpp"
#include "opencv2/stitching/detail/matchers.hpp"

107
modules/stitching/src/timelapsers.cpp Normal file
View File

@ -0,0 +1,107 @@
/*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"
#include "opencl_kernels.hpp"
namespace cv {
namespace detail {
Ptr<Timelapser> Timelapser::createDefault(int type)
{
if (type == AS_IS)
return makePtr<Timelapser>();
if (type == CROP)
return makePtr<TimelapserCrop>();
CV_Error(Error::StsBadArg, "unsupported timelapsing method");
return Ptr<Timelapser>();
}
void Timelapser::initialize(const std::vector<Point> &corners, const std::vector<Size> &sizes)
{
dst_roi_ = resultRoi(corners, sizes);
dst_.create(dst_roi_.size(), CV_16SC3);
}
void Timelapser::process(InputArray _img, InputArray /*_mask*/, Point tl)
{
dst_.setTo(Scalar::all(0));
Mat img = _img.getMat();
Mat dst = dst_.getMat(ACCESS_RW);
CV_Assert(img.type() == CV_16SC3);
int dx = tl.x - dst_roi_.x;
int dy = tl.y - dst_roi_.y;
for (int y = 0; y < img.rows; ++y)
{
const Point3_<short> *src_row = img.ptr<Point3_<short> >(y);
for (int x = 0; x < img.cols; ++x)
{
if (test_point(Point(tl.x + x, tl.y + y)))
{
Point3_<short> *dst_row = dst.ptr<Point3_<short> >(dy + y);
dst_row[dx + x] = src_row[x];
}
}
}
}
bool Timelapser::test_point(Point pt)
{
return dst_roi_.contains(pt);
}
void TimelapserCrop::initialize(const std::vector<Point> &corners, const std::vector<Size> &sizes)
{
dst_roi_ = resultRoiIntersection(corners, sizes);
dst_.create(dst_roi_.size(), CV_16SC3);
}
} // namespace detail
} // namespace cv

15
modules/stitching/src/util.cpp
View File

@ -137,6 +137,21 @@ Rect resultRoi(const std::vector<Point> &corners, const std::vector<Size> &sizes
return Rect(tl, br);
}
Rect resultRoiIntersection(const std::vector<Point> &corners, const std::vector<Size> &sizes)
{
CV_Assert(sizes.size() == corners.size());
Point tl(std::numeric_limits<int>::min(), std::numeric_limits<int>::min());
Point br(std::numeric_limits<int>::max(), std::numeric_limits<int>::max());
for (size_t i = 0; i < corners.size(); ++i)
{
tl.x = std::max(tl.x, corners[i].x);
tl.y = std::max(tl.y, corners[i].y);
br.x = std::min(br.x, corners[i].x + sizes[i].width);
br.y = std::min(br.y, corners[i].y + sizes[i].height);
}
return Rect(tl, br);
}
Point resultTl(const std::vector<Point> &corners)
{

22
modules/videoio/src/cap_gstreamer.cpp
View File

@ -160,13 +160,17 @@ protected:
void CvCapture_GStreamer::init()
{
pipeline = NULL;
frame = NULL;
buffer = NULL;
buffer_caps = NULL;
uridecodebin = NULL;
color = NULL;
sink = NULL;
#if GST_VERSION_MAJOR > 0
sample = NULL;
info = new GstMapInfo;
#endif
buffer = NULL;
caps = NULL;
buffer_caps = NULL;
frame = NULL;
}
/*!
@ -181,31 +185,41 @@ void CvCapture_GStreamer::close()
if(pipeline) {
gst_element_set_state(GST_ELEMENT(pipeline), GST_STATE_NULL);
gst_object_unref(GST_OBJECT(pipeline));
pipeline = NULL;
}
if(uridecodebin){
gst_object_unref(GST_OBJECT(uridecodebin));
uridecodebin = NULL;
}
if(color){
gst_object_unref(GST_OBJECT(color));
color = NULL;
}
if(sink){
gst_object_unref(GST_OBJECT(sink));
sink = NULL;
}
if(buffer)
if(buffer) {
gst_buffer_unref(buffer);
buffer = NULL;
}
if(frame) {
frame->imageData = 0;
cvReleaseImage(&frame);
frame = NULL;
}
if(caps){
gst_caps_unref(caps);
caps = NULL;
}
if(buffer_caps){
gst_caps_unref(buffer_caps);
buffer_caps = NULL;
}
#if GST_VERSION_MAJOR > 0
if(sample){
gst_sample_unref(sample);
sample = NULL;
}
#endif

20
samples/MacOSX/FaceTracker/FaceTracker-Info.plist
View File

@ -1,20 +0,0 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>CFBundleDevelopmentRegion</key>
<string>English</string>
<key>CFBundleExecutable</key>
<string>${EXECUTABLE_NAME}</string>
<key>CFBundleIdentifier</key>
<string>de.rwth-aachen.ient.FaceTracker</string>
<key>CFBundleInfoDictionaryVersion</key>
<string>6.0</string>
<key>CFBundlePackageType</key>
<string>APPL</string>
<key>CFBundleSignature</key>
<string>????</string>
<key>CFBundleVersion</key>
<string>1.0</string>
</dict>
</plist>

86
samples/MacOSX/FaceTracker/FaceTracker.cpp
View File

@ -1,86 +0,0 @@
#include <OpenCV/OpenCV.h>
#include <cassert>
#include <iostream>
const char * WINDOW_NAME = "Face Tracker";
const CFIndex CASCADE_NAME_LEN = 2048;
char CASCADE_NAME[CASCADE_NAME_LEN] = "~/opencv/data/haarcascades/haarcascade_frontalface_alt2.xml";
using namespace std;
int main (int argc, char * const argv[])
{
const int scale = 2;
// locate haar cascade from inside application bundle
// (this is the mac way to package application resources)
CFBundleRef mainBundle = CFBundleGetMainBundle ();
assert (mainBundle);
CFURLRef cascade_url = CFBundleCopyResourceURL (mainBundle, CFSTR("haarcascade_frontalface_alt2"), CFSTR("xml"), NULL);
assert (cascade_url);
Boolean got_it = CFURLGetFileSystemRepresentation (cascade_url, true,
reinterpret_cast<UInt8 *>(CASCADE_NAME), CASCADE_NAME_LEN);
if (! got_it)
abort ();
// create all necessary instances
cvNamedWindow (WINDOW_NAME, CV_WINDOW_AUTOSIZE);
CvCapture * camera = cvCreateCameraCapture (CV_CAP_ANY);
CvHaarClassifierCascade* cascade = (CvHaarClassifierCascade*) cvLoad (CASCADE_NAME, 0, 0, 0);
CvMemStorage* storage = cvCreateMemStorage(0);
assert (storage);
// you do own an iSight, don't you ?!?
if (! camera)
abort ();
// did we load the cascade?!?
if (! cascade)
abort ();
// get an initial frame and duplicate it for later work
IplImage * current_frame = cvQueryFrame (camera);
IplImage * draw_image = cvCreateImage(cvSize (current_frame->width, current_frame->height), IPL_DEPTH_8U, 3);
IplImage * gray_image = cvCreateImage(cvSize (current_frame->width, current_frame->height), IPL_DEPTH_8U, 1);
IplImage * small_image = cvCreateImage(cvSize (current_frame->width / scale, current_frame->height / scale), IPL_DEPTH_8U, 1);
assert (current_frame && gray_image && draw_image);
// as long as there are images ...
while (current_frame = cvQueryFrame (camera))
{
// convert to gray and downsize
cvCvtColor (current_frame, gray_image, CV_BGR2GRAY);
cvResize (gray_image, small_image, CV_INTER_LINEAR);
// detect faces
CvSeq* faces = cvHaarDetectObjects (small_image, cascade, storage,
1.1, 2, CV_HAAR_DO_CANNY_PRUNING,
cvSize (30, 30));
// draw faces
cvFlip (current_frame, draw_image, 1);
for (int i = 0; i < (faces ? faces->total : 0); i++)
{
CvRect* r = (CvRect*) cvGetSeqElem (faces, i);
CvPoint center;
int radius;
center.x = cvRound((small_image->width - r->width*0.5 - r->x) *scale);
center.y = cvRound((r->y + r->height*0.5)*scale);
radius = cvRound((r->width + r->height)*0.25*scale);
cvCircle (draw_image, center, radius, CV_RGB(0,255,0), 3, 8, 0 );
}
// just show the image
cvShowImage (WINDOW_NAME, draw_image);
// wait a tenth of a second for keypress and window drawing
int key = cvWaitKey (100);
if (key == 'q' || key == 'Q')
break;
}
// be nice and return no error
return 0;
}

BIN
samples/MacOSX/FaceTracker/FaceTracker.xcodeproj/project.pbxproj
View File

Binary file not shown.

35
samples/MacOSX/FaceTracker/README.txt
View File

@ -1,35 +0,0 @@
FaceTracker/REAME.txt
2007-05-24, Mark Asbach <asbach@ient.rwth-aachen.de>
Objective:
This document is intended to get you up and running with an OpenCV Framework on Mac OS X
Building the OpenCV.framework:
In the main directory of the opencv distribution, you will find a shell script called
'make_frameworks.sh' that does all of the typical unixy './configure && make' stuff required
to build a universal binary framework. Invoke this script from Terminal.app, wait some minutes
and you are done.
OpenCV is a Private Framework:
On Mac OS X the concept of Framework bundles is meant to simplify distribution of shared libraries,
accompanying headers and documentation. There are however to subtly different 'flavours' of
Frameworks: public and private ones. The public frameworks get installed into the Frameworks
diretories in /Library, /System/Library or ~/Library and are meant to be shared amongst
applications. The private frameworks are only distributed as parts of an Application Bundle.
This makes it easier to deploy applications because they bring their own framework invisibly to
the user. No installation of the framework is necessary and different applications can bring
different versions of the same framework without any conflict.
Since OpenCV is still a moving target, it seems best to avoid any installation and versioning issues
for an end user. The OpenCV framework that currently comes with this demo application therefore
is a Private Framework.
Use it for targets that result in an Application Bundle:
Since it is a Private Framework, it must be copied to the Frameworks/ directory of an Application
Bundle, which means, it is useless for plain unix console applications. You should create a Carbon
or a Cocoa application target in XCode for your projects. Then add the OpenCV.framework just like
in this demo and add a Copy Files build phase to your target. Let that phase copy to the Framework
directory and drop the OpenCV.framework on the build phase (again just like in this demo code).
The resulting application bundle will be self contained and if you set compiler option correctly
(in the "Build" tab of the "Project Info" window you should find 'i386 ppc' for the architectures),
your application can just be copied to any OS 10.4 Mac and used without further installation.

11
samples/cpp/H1to3p.xml Executable file
View File

@ -0,0 +1,11 @@
<?xml version="1.0"?>
<opencv_storage>
<H13 type_id="opencv-matrix">
<rows>3</rows>
<cols>3</cols>
<dt>d</dt>
<data>
7.6285898e-01 -2.9922929e-01 2.2567123e+02
3.3443473e-01 1.0143901e+00 -7.6999973e+01
3.4663091e-04 -1.4364524e-05 1.0000000e+00 </data></H13>
</opencv_storage>

BIN
samples/cpp/graf1.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 929 KiB

BIN
samples/cpp/graf3.png Executable file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 953 KiB

705
samples/cpp/linemod.cpp
View File

@ -1,705 +0,0 @@
#include <opencv2/core.hpp>
#include <opencv2/core/utility.hpp>
#include <opencv2/imgproc/imgproc_c.h> // cvFindContours
#include <opencv2/imgproc.hpp>
#include <opencv2/objdetect.hpp>
#include <opencv2/videoio.hpp>
#include <opencv2/highgui.hpp>
#include <iterator>
#include <set>
#include <cstdio>
#include <iostream>
// Function prototypes
void subtractPlane(const cv::Mat& depth, cv::Mat& mask, std::vector<CvPoint>& chain, double f);
std::vector<CvPoint> maskFromTemplate(const std::vector<cv::linemod::Template>& templates,
int num_modalities, cv::Point offset, cv::Size size,
cv::Mat& mask, cv::Mat& dst);
void templateConvexHull(const std::vector<cv::linemod::Template>& templates,
int num_modalities, cv::Point offset, cv::Size size,
cv::Mat& dst);
void drawResponse(const std::vector<cv::linemod::Template>& templates,
int num_modalities, cv::Mat& dst, cv::Point offset, int T);
cv::Mat displayQuantized(const cv::Mat& quantized);
// Copy of cv_mouse from cv_utilities
class Mouse
{
public:
static void start(const std::string& a_img_name)
{
cv::setMouseCallback(a_img_name.c_str(), Mouse::cv_on_mouse, 0);
}
static int event(void)
{
int l_event = m_event;
m_event = -1;
return l_event;
}
static int x(void)
{
return m_x;
}
static int y(void)
{
return m_y;
}
private:
static void cv_on_mouse(int a_event, int a_x, int a_y, int, void *)
{
m_event = a_event;
m_x = a_x;
m_y = a_y;
}
static int m_event;
static int m_x;
static int m_y;
};
int Mouse::m_event;
int Mouse::m_x;
int Mouse::m_y;
static void help()
{
printf("Usage: openni_demo [templates.yml]\n\n"
"Place your object on a planar, featureless surface. With the mouse,\n"
"frame it in the 'color' window and right click to learn a first template.\n"
"Then press 'l' to enter online learning mode, and move the camera around.\n"
"When the match score falls between 90-95%% the demo will add a new template.\n\n"
"Keys:\n"
"\t h -- This help page\n"
"\t l -- Toggle online learning\n"
"\t m -- Toggle printing match result\n"
"\t t -- Toggle printing timings\n"
"\t w -- Write learned templates to disk\n"
"\t [ ] -- Adjust matching threshold: '[' down, ']' up\n"
"\t q -- Quit\n\n");
}
// Adapted from cv_timer in cv_utilities
class Timer
{
public:
Timer() : start_(0), time_(0) {}
void start()
{
start_ = cv::getTickCount();
}
void stop()
{
CV_Assert(start_ != 0);
int64 end = cv::getTickCount();
time_ += end - start_;
start_ = 0;
}
double time()
{
double ret = time_ / cv::getTickFrequency();
time_ = 0;
return ret;
}
private:
int64 start_, time_;
};
// Functions to store detector and templates in single XML/YAML file
static cv::Ptr<cv::linemod::Detector> readLinemod(const std::string& filename)
{
cv::Ptr<cv::linemod::Detector> detector = cv::makePtr<cv::linemod::Detector>();
cv::FileStorage fs(filename, cv::FileStorage::READ);
detector->read(fs.root());
cv::FileNode fn = fs["classes"];
for (cv::FileNodeIterator i = fn.begin(), iend = fn.end(); i != iend; ++i)
detector->readClass(*i);
return detector;
}
static void writeLinemod(const cv::Ptr<cv::linemod::Detector>& detector, const std::string& filename)
{
cv::FileStorage fs(filename, cv::FileStorage::WRITE);
detector->write(fs);
std::vector<cv::String> ids = detector->classIds();
fs << "classes" << "[";
for (int i = 0; i < (int)ids.size(); ++i)
{
fs << "{";
detector->writeClass(ids[i], fs);
fs << "}"; // current class
}
fs << "]"; // classes
}
int main(int argc, char * argv[])
{
// Various settings and flags
bool show_match_result = true;
bool show_timings = false;
bool learn_online = false;
int num_classes = 0;
int matching_threshold = 80;
/// @todo Keys for changing these?
cv::Size roi_size(200, 200);
int learning_lower_bound = 90;
int learning_upper_bound = 95;
// Timers
Timer extract_timer;
Timer match_timer;
// Initialize HighGUI
help();
cv::namedWindow("color");
cv::namedWindow("normals");
Mouse::start("color");
// Initialize LINEMOD data structures
cv::Ptr<cv::linemod::Detector> detector;
std::string filename;
if (argc == 1)
{
filename = "linemod_templates.yml";
detector = cv::linemod::getDefaultLINEMOD();
}
else
{
detector = readLinemod(argv[1]);
std::vector<cv::String> ids = detector->classIds();
num_classes = detector->numClasses();
printf("Loaded %s with %d classes and %d templates\n",
argv[1], num_classes, detector->numTemplates());
if (!ids.empty())
{
printf("Class ids:\n");
std::copy(ids.begin(), ids.end(), std::ostream_iterator<std::string>(std::cout, "\n"));
}
}
int num_modalities = (int)detector->getModalities().size();
// Open Kinect sensor
cv::VideoCapture capture( cv::CAP_OPENNI );
if (!capture.isOpened())
{
printf("Could not open OpenNI-capable sensor\n");
return -1;
}
capture.set(cv::CAP_PROP_OPENNI_REGISTRATION, 1);
double focal_length = capture.get(cv::CAP_OPENNI_DEPTH_GENERATOR_FOCAL_LENGTH);
//printf("Focal length = %f\n", focal_length);
// Main loop
cv::Mat color, depth;
for(;;)
{
// Capture next color/depth pair
capture.grab();
capture.retrieve(depth, cv::CAP_OPENNI_DEPTH_MAP);
capture.retrieve(color, cv::CAP_OPENNI_BGR_IMAGE);
std::vector<cv::Mat> sources;
sources.push_back(color);
sources.push_back(depth);
cv::Mat display = color.clone();
if (!learn_online)
{
cv::Point mouse(Mouse::x(), Mouse::y());
int event = Mouse::event();
// Compute ROI centered on current mouse location
cv::Point roi_offset(roi_size.width / 2, roi_size.height / 2);
cv::Point pt1 = mouse - roi_offset; // top left
cv::Point pt2 = mouse + roi_offset; // bottom right
if (event == cv::EVENT_RBUTTONDOWN)
{
// Compute object mask by subtracting the plane within the ROI
std::vector<CvPoint> chain(4);
chain[0] = pt1;
chain[1] = cv::Point(pt2.x, pt1.y);
chain[2] = pt2;
chain[3] = cv::Point(pt1.x, pt2.y);
cv::Mat mask;
subtractPlane(depth, mask, chain, focal_length);
cv::imshow("mask", mask);
// Extract template
std::string class_id = cv::format("class%d", num_classes);
cv::Rect bb;
extract_timer.start();
int template_id = detector->addTemplate(sources, class_id, mask, &bb);
extract_timer.stop();
if (template_id != -1)
{
printf("*** Added template (id %d) for new object class %d***\n",
template_id, num_classes);
//printf("Extracted at (%d, %d) size %dx%d\n", bb.x, bb.y, bb.width, bb.height);
}
++num_classes;
}
// Draw ROI for display
cv::rectangle(display, pt1, pt2, CV_RGB(0,0,0), 3);
cv::rectangle(display, pt1, pt2, CV_RGB(255,255,0), 1);
}
// Perform matching
std::vector<cv::linemod::Match> matches;
std::vector<cv::String> class_ids;
std::vector<cv::Mat> quantized_images;
match_timer.start();
detector->match(sources, (float)matching_threshold, matches, class_ids, quantized_images);
match_timer.stop();
int classes_visited = 0;
std::set<std::string> visited;
for (int i = 0; (i < (int)matches.size()) && (classes_visited < num_classes); ++i)
{
cv::linemod::Match m = matches[i];
if (visited.insert(m.class_id).second)
{
++classes_visited;
if (show_match_result)
{
printf("Similarity: %5.1f%%; x: %3d; y: %3d; class: %s; template: %3d\n",
m.similarity, m.x, m.y, m.class_id.c_str(), m.template_id);
}
// Draw matching template
const std::vector<cv::linemod::Template>& templates = detector->getTemplates(m.class_id, m.template_id);
drawResponse(templates, num_modalities, display, cv::Point(m.x, m.y), detector->getT(0));
if (learn_online == true)
{
/// @todo Online learning possibly broken by new gradient feature extraction,
/// which assumes an accurate object outline.
// Compute masks based on convex hull of matched template
cv::Mat color_mask, depth_mask;
std::vector<CvPoint> chain = maskFromTemplate(templates, num_modalities,
cv::Point(m.x, m.y), color.size(),
color_mask, display);
subtractPlane(depth, depth_mask, chain, focal_length);
cv::imshow("mask", depth_mask);
// If pretty sure (but not TOO sure), add new template
if (learning_lower_bound < m.similarity && m.similarity < learning_upper_bound)
{
extract_timer.start();
int template_id = detector->addTemplate(sources, m.class_id, depth_mask);
extract_timer.stop();
if (template_id != -1)
{
printf("*** Added template (id %d) for existing object class %s***\n",
template_id, m.class_id.c_str());
}
}
}
}
}
if (show_match_result && matches.empty())
printf("No matches found...\n");
if (show_timings)
{
printf("Training: %.2fs\n", extract_timer.time());
printf("Matching: %.2fs\n", match_timer.time());
}
if (show_match_result || show_timings)
printf("------------------------------------------------------------\n");
cv::imshow("color", display);
cv::imshow("normals", quantized_images[1]);
cv::FileStorage fs;
char key = (char)cv::waitKey(10);
if( key == 'q' )
break;
switch (key)
{
case 'h':
help();
break;
case 'm':
// toggle printing match result
show_match_result = !show_match_result;
printf("Show match result %s\n", show_match_result ? "ON" : "OFF");
break;
case 't':
// toggle printing timings
show_timings = !show_timings;
printf("Show timings %s\n", show_timings ? "ON" : "OFF");
break;
case 'l':
// toggle online learning
learn_online = !learn_online;
printf("Online learning %s\n", learn_online ? "ON" : "OFF");
break;
case '[':
// decrement threshold
matching_threshold = std::max(matching_threshold - 1, -100);
printf("New threshold: %d\n", matching_threshold);
break;
case ']':
// increment threshold
matching_threshold = std::min(matching_threshold + 1, +100);
printf("New threshold: %d\n", matching_threshold);
break;
case 'w':
// write model to disk
writeLinemod(detector, filename);
printf("Wrote detector and templates to %s\n", filename.c_str());
break;
default:
;
}
}
return 0;
}
static void reprojectPoints(const std::vector<cv::Point3d>& proj, std::vector<cv::Point3d>& real, double f)
{
real.resize(proj.size());
double f_inv = 1.0 / f;
for (int i = 0; i < (int)proj.size(); ++i)
{
double Z = proj[i].z;
real[i].x = (proj[i].x - 320.) * (f_inv * Z);
real[i].y = (proj[i].y - 240.) * (f_inv * Z);
real[i].z = Z;
}
}
static void filterPlane(IplImage * ap_depth, std::vector<IplImage *> & a_masks, std::vector<CvPoint> & a_chain, double f)
{
const int l_num_cost_pts = 200;
float l_thres = 4;
IplImage * lp_mask = cvCreateImage(cvGetSize(ap_depth), IPL_DEPTH_8U, 1);
cvSet(lp_mask, cvRealScalar(0));
std::vector<CvPoint> l_chain_vector;
float l_chain_length = 0;
float * lp_seg_length = new float[a_chain.size()];
for (int l_i = 0; l_i < (int)a_chain.size(); ++l_i)
{
float x_diff = (float)(a_chain[(l_i + 1) % a_chain.size()].x - a_chain[l_i].x);
float y_diff = (float)(a_chain[(l_i + 1) % a_chain.size()].y - a_chain[l_i].y);
lp_seg_length[l_i] = sqrt(x_diff*x_diff + y_diff*y_diff);
l_chain_length += lp_seg_length[l_i];
}
for (int l_i = 0; l_i < (int)a_chain.size(); ++l_i)
{
if (lp_seg_length[l_i] > 0)
{
int l_cur_num = cvRound(l_num_cost_pts * lp_seg_length[l_i] / l_chain_length);
float l_cur_len = lp_seg_length[l_i] / l_cur_num;
for (int l_j = 0; l_j < l_cur_num; ++l_j)
{
float l_ratio = (l_cur_len * l_j / lp_seg_length[l_i]);
CvPoint l_pts;
l_pts.x = cvRound(l_ratio * (a_chain[(l_i + 1) % a_chain.size()].x - a_chain[l_i].x) + a_chain[l_i].x);
l_pts.y = cvRound(l_ratio * (a_chain[(l_i + 1) % a_chain.size()].y - a_chain[l_i].y) + a_chain[l_i].y);
l_chain_vector.push_back(l_pts);
}
}
}
std::vector<cv::Point3d> lp_src_3Dpts(l_chain_vector.size());
for (int l_i = 0; l_i < (int)l_chain_vector.size(); ++l_i)
{
lp_src_3Dpts[l_i].x = l_chain_vector[l_i].x;
lp_src_3Dpts[l_i].y = l_chain_vector[l_i].y;
lp_src_3Dpts[l_i].z = CV_IMAGE_ELEM(ap_depth, unsigned short, cvRound(lp_src_3Dpts[l_i].y), cvRound(lp_src_3Dpts[l_i].x));
//CV_IMAGE_ELEM(lp_mask,unsigned char,(int)lp_src_3Dpts[l_i].Y,(int)lp_src_3Dpts[l_i].X)=255;
}
//cv_show_image(lp_mask,"hallo2");
reprojectPoints(lp_src_3Dpts, lp_src_3Dpts, f);
CvMat * lp_pts = cvCreateMat((int)l_chain_vector.size(), 4, CV_32F);
CvMat * lp_v = cvCreateMat(4, 4, CV_32F);
CvMat * lp_w = cvCreateMat(4, 1, CV_32F);
for (int l_i = 0; l_i < (int)l_chain_vector.size(); ++l_i)
{
CV_MAT_ELEM(*lp_pts, float, l_i, 0) = (float)lp_src_3Dpts[l_i].x;
CV_MAT_ELEM(*lp_pts, float, l_i, 1) = (float)lp_src_3Dpts[l_i].y;
CV_MAT_ELEM(*lp_pts, float, l_i, 2) = (float)lp_src_3Dpts[l_i].z;
CV_MAT_ELEM(*lp_pts, float, l_i, 3) = 1.0f;
}
cvSVD(lp_pts, lp_w, 0, lp_v);
float l_n[4] = {CV_MAT_ELEM(*lp_v, float, 0, 3),
CV_MAT_ELEM(*lp_v, float, 1, 3),
CV_MAT_ELEM(*lp_v, float, 2, 3),
CV_MAT_ELEM(*lp_v, float, 3, 3)};
float l_norm = sqrt(l_n[0] * l_n[0] + l_n[1] * l_n[1] + l_n[2] * l_n[2]);
l_n[0] /= l_norm;
l_n[1] /= l_norm;
l_n[2] /= l_norm;
l_n[3] /= l_norm;
float l_max_dist = 0;
for (int l_i = 0; l_i < (int)l_chain_vector.size(); ++l_i)
{
float l_dist = l_n[0] * CV_MAT_ELEM(*lp_pts, float, l_i, 0) +
l_n[1] * CV_MAT_ELEM(*lp_pts, float, l_i, 1) +
l_n[2] * CV_MAT_ELEM(*lp_pts, float, l_i, 2) +
l_n[3] * CV_MAT_ELEM(*lp_pts, float, l_i, 3);
if (fabs(l_dist) > l_max_dist)
l_max_dist = l_dist;
}
//std::cerr << "plane: " << l_n[0] << ";" << l_n[1] << ";" << l_n[2] << ";" << l_n[3] << " maxdist: " << l_max_dist << " end" << std::endl;
int l_minx = ap_depth->width;
int l_miny = ap_depth->height;
int l_maxx = 0;
int l_maxy = 0;
for (int l_i = 0; l_i < (int)a_chain.size(); ++l_i)
{
l_minx = std::min(l_minx, a_chain[l_i].x);
l_miny = std::min(l_miny, a_chain[l_i].y);
l_maxx = std::max(l_maxx, a_chain[l_i].x);
l_maxy = std::max(l_maxy, a_chain[l_i].y);
}
int l_w = l_maxx - l_minx + 1;
int l_h = l_maxy - l_miny + 1;
int l_nn = (int)a_chain.size();
CvPoint * lp_chain = new CvPoint[l_nn];
for (int l_i = 0; l_i < l_nn; ++l_i)
lp_chain[l_i] = a_chain[l_i];
cvFillPoly(lp_mask, &lp_chain, &l_nn, 1, cvScalar(255, 255, 255));
delete[] lp_chain;
//cv_show_image(lp_mask,"hallo1");
std::vector<cv::Point3d> lp_dst_3Dpts(l_h * l_w);
int l_ind = 0;
for (int l_r = 0; l_r < l_h; ++l_r)
{
for (int l_c = 0; l_c < l_w; ++l_c)
{
lp_dst_3Dpts[l_ind].x = l_c + l_minx;
lp_dst_3Dpts[l_ind].y = l_r + l_miny;
lp_dst_3Dpts[l_ind].z = CV_IMAGE_ELEM(ap_depth, unsigned short, l_r + l_miny, l_c + l_minx);
++l_ind;
}
}
reprojectPoints(lp_dst_3Dpts, lp_dst_3Dpts, f);
l_ind = 0;
for (int l_r = 0; l_r < l_h; ++l_r)
{
for (int l_c = 0; l_c < l_w; ++l_c)
{
float l_dist = (float)(l_n[0] * lp_dst_3Dpts[l_ind].x + l_n[1] * lp_dst_3Dpts[l_ind].y + lp_dst_3Dpts[l_ind].z * l_n[2] + l_n[3]);
++l_ind;
if (CV_IMAGE_ELEM(lp_mask, unsigned char, l_r + l_miny, l_c + l_minx) != 0)
{
if (fabs(l_dist) < std::max(l_thres, (l_max_dist * 2.0f)))
{
for (int l_p = 0; l_p < (int)a_masks.size(); ++l_p)
{
int l_col = cvRound((l_c + l_minx) / (l_p + 1.0));
int l_row = cvRound((l_r + l_miny) / (l_p + 1.0));
CV_IMAGE_ELEM(a_masks[l_p], unsigned char, l_row, l_col) = 0;
}
}
else
{
for (int l_p = 0; l_p < (int)a_masks.size(); ++l_p)
{
int l_col = cvRound((l_c + l_minx) / (l_p + 1.0));
int l_row = cvRound((l_r + l_miny) / (l_p + 1.0));
CV_IMAGE_ELEM(a_masks[l_p], unsigned char, l_row, l_col) = 255;
}
}
}
}
}
cvReleaseImage(&lp_mask);
cvReleaseMat(&lp_pts);
cvReleaseMat(&lp_w);
cvReleaseMat(&lp_v);
}
void subtractPlane(const cv::Mat& depth, cv::Mat& mask, std::vector<CvPoint>& chain, double f)
{
mask = cv::Mat::zeros(depth.size(), CV_8U);
std::vector<IplImage*> tmp;
IplImage mask_ipl = mask;
tmp.push_back(&mask_ipl);
IplImage depth_ipl = depth;
filterPlane(&depth_ipl, tmp, chain, f);
}
std::vector<CvPoint> maskFromTemplate(const std::vector<cv::linemod::Template>& templates,
int num_modalities, cv::Point offset, cv::Size size,
cv::Mat& mask, cv::Mat& dst)
{
templateConvexHull(templates, num_modalities, offset, size, mask);
const int OFFSET = 30;
cv::dilate(mask, mask, cv::Mat(), cv::Point(-1,-1), OFFSET);
CvMemStorage * lp_storage = cvCreateMemStorage(0);
CvTreeNodeIterator l_iterator;
CvSeqReader l_reader;
CvSeq * lp_contour = 0;
cv::Mat mask_copy = mask.clone();
IplImage mask_copy_ipl = mask_copy;
cvFindContours(&mask_copy_ipl, lp_storage, &lp_contour, sizeof(CvContour),
CV_RETR_CCOMP, CV_CHAIN_APPROX_SIMPLE);
std::vector<CvPoint> l_pts1; // to use as input to cv_primesensor::filter_plane
cvInitTreeNodeIterator(&l_iterator, lp_contour, 1);
while ((lp_contour = (CvSeq *)cvNextTreeNode(&l_iterator)) != 0)
{
CvPoint l_pt0;
cvStartReadSeq(lp_contour, &l_reader, 0);
CV_READ_SEQ_ELEM(l_pt0, l_reader);
l_pts1.push_back(l_pt0);
for (int i = 0; i < lp_contour->total; ++i)
{
CvPoint l_pt1;
CV_READ_SEQ_ELEM(l_pt1, l_reader);
/// @todo Really need dst at all? Can just as well do this outside
cv::line(dst, l_pt0, l_pt1, CV_RGB(0, 255, 0), 2);
l_pt0 = l_pt1;
l_pts1.push_back(l_pt0);
}
}
cvReleaseMemStorage(&lp_storage);
return l_pts1;
}
// Adapted from cv_show_angles
cv::Mat displayQuantized(const cv::Mat& quantized)
{
cv::Mat color(quantized.size(), CV_8UC3);
for (int r = 0; r < quantized.rows; ++r)
{
const uchar* quant_r = quantized.ptr(r);
cv::Vec3b* color_r = color.ptr<cv::Vec3b>(r);
for (int c = 0; c < quantized.cols; ++c)
{
cv::Vec3b& bgr = color_r[c];
switch (quant_r[c])
{
case 0: bgr[0]= 0; bgr[1]= 0; bgr[2]= 0; break;
case 1: bgr[0]= 55; bgr[1]= 55; bgr[2]= 55; break;
case 2: bgr[0]= 80; bgr[1]= 80; bgr[2]= 80; break;
case 4: bgr[0]=105; bgr[1]=105; bgr[2]=105; break;
case 8: bgr[0]=130; bgr[1]=130; bgr[2]=130; break;
case 16: bgr[0]=155; bgr[1]=155; bgr[2]=155; break;
case 32: bgr[0]=180; bgr[1]=180; bgr[2]=180; break;
case 64: bgr[0]=205; bgr[1]=205; bgr[2]=205; break;
case 128: bgr[0]=230; bgr[1]=230; bgr[2]=230; break;
case 255: bgr[0]= 0; bgr[1]= 0; bgr[2]=255; break;
default: bgr[0]= 0; bgr[1]=255; bgr[2]= 0; break;
}
}
}
return color;
}
// Adapted from cv_line_template::convex_hull
void templateConvexHull(const std::vector<cv::linemod::Template>& templates,
int num_modalities, cv::Point offset, cv::Size size,
cv::Mat& dst)
{
std::vector<cv::Point> points;
for (int m = 0; m < num_modalities; ++m)
{
for (int i = 0; i < (int)templates[m].features.size(); ++i)
{
cv::linemod::Feature f = templates[m].features[i];
points.push_back(cv::Point(f.x, f.y) + offset);
}
}
std::vector<cv::Point> hull;
cv::convexHull(points, hull);
dst = cv::Mat::zeros(size, CV_8U);
const int hull_count = (int)hull.size();
const cv::Point* hull_pts = &hull[0];
cv::fillPoly(dst, &hull_pts, &hull_count, 1, cv::Scalar(255));
}
void drawResponse(const std::vector<cv::linemod::Template>& templates,
int num_modalities, cv::Mat& dst, cv::Point offset, int T)
{
static const cv::Scalar COLORS[5] = { CV_RGB(0, 0, 255),
CV_RGB(0, 255, 0),
CV_RGB(255, 255, 0),
CV_RGB(255, 140, 0),
CV_RGB(255, 0, 0) };
for (int m = 0; m < num_modalities; ++m)
{
// NOTE: Original demo recalculated max response for each feature in the TxT
// box around it and chose the display color based on that response. Here
// the display color just depends on the modality.
cv::Scalar color = COLORS[m];
for (int i = 0; i < (int)templates[m].features.size(); ++i)
{
cv::linemod::Feature f = templates[m].features[i];
cv::Point pt(f.x + offset.x, f.y + offset.y);
cv::circle(dst, pt, T / 2, color);
}
}
}

BIN
samples/cpp/scenetext01.jpg
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 95 KiB

BIN
samples/cpp/scenetext02.jpg
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 93 KiB

BIN
samples/cpp/scenetext03.jpg
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 59 KiB

BIN
samples/cpp/scenetext04.jpg
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 97 KiB

BIN
samples/cpp/scenetext05.jpg
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 111 KiB

BIN
samples/cpp/scenetext06.jpg
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 69 KiB

74
samples/cpp/stitching_detailed.cpp
View File

@ -50,6 +50,7 @@
#include "opencv2/highgui.hpp"
#include "opencv2/stitching/detail/autocalib.hpp"
#include "opencv2/stitching/detail/blenders.hpp"
#include "opencv2/stitching/detail/timelapsers.hpp"
#include "opencv2/stitching/detail/camera.hpp"
#include "opencv2/stitching/detail/exposure_compensate.hpp"
#include "opencv2/stitching/detail/matchers.hpp"
@ -117,7 +118,9 @@ static void printUsage()
" --blend_strength <float>\n"
" Blending strength from [0,100] range. The default is 5.\n"
" --output <result_img>\n"
" The default is 'result.jpg'.\n";
" The default is 'result.jpg'.\n"
" --timelapse (as_is|crop) (range_width)\n"
" Output warped images separately as frames of a time lapse movie, with 'fixed_' prepended to input file names.\n";
}
@ -141,8 +144,12 @@ int expos_comp_type = ExposureCompensator::GAIN_BLOCKS;
float match_conf = 0.3f;
string seam_find_type = "gc_color";
int blend_type = Blender::MULTI_BAND;
int timelapse_type = Timelapser::AS_IS;
float blend_strength = 5;
string result_name = "result.jpg";
bool timelapse = false;
int timelapse_range = 5;
static int parseCmdArgs(int argc, char** argv)
{
@ -305,6 +312,24 @@ static int parseCmdArgs(int argc, char** argv)
}
i++;
}
else if (string(argv[i]) == "--timelapse")
{
timelapse = true;
if (string(argv[i + 1]) == "as_is")
timelapse_type = Timelapser::AS_IS;
else if (string(argv[i + 1]) == "crop")
timelapse_type = Timelapser::CROP;
else
{
cout << "Bad timelapse method\n";
return -1;
}
i++;
timelapse_range = atoi(argv[i + 1]);
i++;
}
else if (string(argv[i]) == "--blend_strength")
{
blend_strength = static_cast<float>(atof(argv[i + 1]));
@ -433,9 +458,19 @@ int main(int argc, char* argv[])
t = getTickCount();
#endif
vector<MatchesInfo> pairwise_matches;
BestOf2NearestMatcher matcher(try_cuda, match_conf);
matcher(features, pairwise_matches);
matcher.collectGarbage();
if (!timelapse)
{
BestOf2NearestMatcher matcher(try_cuda, match_conf);
matcher(features, pairwise_matches);
matcher.collectGarbage();
}
else
{
BestOf2NearestRangeMatcher matcher(timelapse_range, try_cuda, match_conf);
matcher(features, pairwise_matches);
matcher.collectGarbage();
}
LOGLN("Pairwise matching, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec");
// Check if we should save matches graph
@ -680,6 +715,7 @@ int main(int argc, char* argv[])
Mat img_warped, img_warped_s;
Mat dilated_mask, seam_mask, mask, mask_warped;
Ptr<Blender> blender;
Ptr<Timelapser> timelapser;
//double compose_seam_aspect = 1;
double compose_work_aspect = 1;
@ -756,7 +792,7 @@ int main(int argc, char* argv[])
resize(dilated_mask, seam_mask, mask_warped.size());
mask_warped = seam_mask & mask_warped;
if (!blender)
if (!blender && !timelapse)
{
blender = Blender::createDefault(blend_type, try_cuda);
Size dst_sz = resultRoi(corners, sizes).size();
@ -777,17 +813,35 @@ int main(int argc, char* argv[])
}
blender->prepare(corners, sizes);
}
else if (!timelapser)
{
CV_Assert(timelapse);
timelapser = Timelapser::createDefault(timelapse_type);
timelapser->initialize(corners, sizes);
}
// Blend the current image
blender->feed(img_warped_s, mask_warped, corners[img_idx]);
if (timelapse)
{
timelapser->process(img_warped_s, Mat::ones(img_warped_s.size(), CV_8UC1), corners[img_idx]);
imwrite("fixed_" + img_names[img_idx], timelapser->getDst());
}
else
{
blender->feed(img_warped_s, mask_warped, corners[img_idx]);
}
}
Mat result, result_mask;
blender->blend(result, result_mask);
if (!timelapse)
{
Mat result, result_mask;
blender->blend(result, result_mask);
LOGLN("Compositing, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec");
LOGLN("Compositing, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec");
imwrite(result_name, result);
imwrite(result_name, result);
}
LOGLN("Finished, total time: " << ((getTickCount() - app_start_time) / getTickFrequency()) << " sec");
return 0;

128
samples/cpp/textdetection.cpp
View File

@ -1,128 +0,0 @@
/*
* textdetection.cpp
*
* A demo program of the Extremal Region Filter algorithm described in
* Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012
*
* Created on: Sep 23, 2013
* Author: Lluis Gomez i Bigorda <lgomez AT cvc.uab.es>
*/
#include "opencv2/opencv.hpp"
#include "opencv2/objdetect.hpp"
#include "opencv2/imgcodecs.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/imgproc.hpp"
#include <vector>
#include <iostream>
#include <iomanip>
using namespace std;
using namespace cv;
void show_help_and_exit(const char *cmd);
void groups_draw(Mat &src, vector<Rect> &groups);
void er_show(vector<Mat> &channels, vector<vector<ERStat> > &regions);
int main(int argc, const char * argv[])
{
cout << endl << argv[0] << endl << endl;
cout << "Demo program of the Extremal Region Filter algorithm described in " << endl;
cout << "Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012" << endl << endl;
if (argc < 2) show_help_and_exit(argv[0]);
Mat src = imread(argv[1]);
// Extract channels to be processed individually
vector<Mat> channels;
computeNMChannels(src, channels);
int cn = (int)channels.size();
// Append negative channels to detect ER- (bright regions over dark background)
for (int c = 0; c < cn-1; c++)
channels.push_back(255-channels[c]);
// Create ERFilter objects with the 1st and 2nd stage default classifiers
Ptr<ERFilter> er_filter1 = createERFilterNM1(loadClassifierNM1("trained_classifierNM1.xml"),16,0.00015f,0.13f,0.2f,true,0.1f);
Ptr<ERFilter> er_filter2 = createERFilterNM2(loadClassifierNM2("trained_classifierNM2.xml"),0.5);
vector<vector<ERStat> > regions(channels.size());
// Apply the default cascade classifier to each independent channel (could be done in parallel)
cout << "Extracting Class Specific Extremal Regions from " << (int)channels.size() << " channels ..." << endl;
cout << " (...) this may take a while (...)" << endl << endl;
for (int c=0; c<(int)channels.size(); c++)
{
er_filter1->run(channels[c], regions[c]);
er_filter2->run(channels[c], regions[c]);
}
// Detect character groups
cout << "Grouping extracted ERs ... ";
vector<Rect> groups;
erGrouping(channels, regions, "trained_classifier_erGrouping.xml", 0.5, groups);
// draw groups
groups_draw(src, groups);
imshow("grouping",src);
cout << "Done!" << endl << endl;
cout << "Press 'e' to show the extracted Extremal Regions, any other key to exit." << endl << endl;
if( waitKey (-1) == 101)
er_show(channels,regions);
// memory clean-up
er_filter1.release();
er_filter2.release();
regions.clear();
if (!groups.empty())
{
groups.clear();
}
}
// helper functions
void show_help_and_exit(const char *cmd)
{
cout << " Usage: " << cmd << " <input_image> " << endl;
cout << " Default classifier files (trained_classifierNM*.xml) must be in current directory" << endl << endl;
exit(-1);
}
void groups_draw(Mat &src, vector<Rect> &groups)
{
for (int i=(int)groups.size()-1; i>=0; i--)
{
if (src.type() == CV_8UC3)
rectangle(src,groups.at(i).tl(),groups.at(i).br(),Scalar( 0, 255, 255 ), 3, 8 );
else
rectangle(src,groups.at(i).tl(),groups.at(i).br(),Scalar( 255 ), 3, 8 );
}
}
void er_show(vector<Mat> &channels, vector<vector<ERStat> > &regions)
{
for (int c=0; c<(int)channels.size(); c++)
{
Mat dst = Mat::zeros(channels[0].rows+2,channels[0].cols+2,CV_8UC1);
for (int r=0; r<(int)regions[c].size(); r++)
{
ERStat er = regions[c][r];
if (er.parent != NULL) // deprecate the root region
{
int newMaskVal = 255;
int flags = 4 + (newMaskVal << 8) + FLOODFILL_FIXED_RANGE + FLOODFILL_MASK_ONLY;
floodFill(channels[c],dst,Point(er.pixel%channels[c].cols,er.pixel/channels[c].cols),
Scalar(255),0,Scalar(er.level),Scalar(0),flags);
}
}
char buff[10]; char *buff_ptr = buff;
sprintf(buff, "channel %d", c);
imshow(buff_ptr, dst);
}
waitKey(-1);
}

79
samples/cpp/tutorial_code/features2D/AKAZE_match.cpp Executable file
View File

@ -0,0 +1,79 @@
#include <opencv2/features2d.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/opencv.hpp>
#include <vector>
#include <iostream>
using namespace std;
using namespace cv;
const float inlier_threshold = 2.5f; // Distance threshold to identify inliers
const float nn_match_ratio = 0.8f; // Nearest neighbor matching ratio
int main(void)
{
Mat img1 = imread("graf1.png", IMREAD_GRAYSCALE);
Mat img2 = imread("graf3.png", IMREAD_GRAYSCALE);
Mat homography;
FileStorage fs("H1to3p.xml", FileStorage::READ);
fs.getFirstTopLevelNode() >> homography;
vector<KeyPoint> kpts1, kpts2;
Mat desc1, desc2;
AKAZE akaze;
akaze(img1, noArray(), kpts1, desc1);
akaze(img2, noArray(), kpts2, desc2);
BFMatcher matcher(NORM_HAMMING);
vector< vector<DMatch> > nn_matches;
matcher.knnMatch(desc1, desc2, nn_matches, 2);
vector<KeyPoint> matched1, matched2, inliers1, inliers2;
vector<DMatch> good_matches;
for(size_t i = 0; i < nn_matches.size(); i++) {
DMatch first = nn_matches[i][0];
float dist1 = nn_matches[i][0].distance;
float dist2 = nn_matches[i][1].distance;
if(dist1 < nn_match_ratio * dist2) {
matched1.push_back(kpts1[first.queryIdx]);
matched2.push_back(kpts2[first.trainIdx]);
}
}
for(unsigned i = 0; i < matched1.size(); i++) {
Mat col = Mat::ones(3, 1, CV_64F);
col.at<double>(0) = matched1[i].pt.x;
col.at<double>(1) = matched1[i].pt.y;
col = homography * col;
col /= col.at<double>(2);
double dist = sqrt( pow(col.at<double>(0) - matched2[i].pt.x, 2) +
pow(col.at<double>(1) - matched2[i].pt.y, 2));
if(dist < inlier_threshold) {
int new_i = static_cast<int>(inliers1.size());
inliers1.push_back(matched1[i]);
inliers2.push_back(matched2[i]);
good_matches.push_back(DMatch(new_i, new_i, 0));
}
}
Mat res;
drawMatches(img1, inliers1, img2, inliers2, good_matches, res);
imwrite("res.png", res);
double inlier_ratio = inliers1.size() * 1.0 / matched1.size();
cout << "A-KAZE Matching Results" << endl;
cout << "*******************************" << endl;
cout << "# Keypoints 1: \t" << kpts1.size() << endl;
cout << "# Keypoints 2: \t" << kpts2.size() << endl;
cout << "# Matches: \t" << matched1.size() << endl;
cout << "# Inliers: \t" << inliers1.size() << endl;
cout << "# Inliers Ratio: \t" << inlier_ratio << endl;
cout << endl;
return 0;
}