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

Merge pull request #1061 from bitwangyaoyao:2.4_fix2

Browse Source
This commit is contained in:
Roman Donchenko 2013-07-01 13:21:12 +04:00 committed by OpenCV Buildbot
commit c8cd2cf601
10 changed files with 186 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
modules/ocl/src/hog.cpp
View File

@ -1816,8 +1816,14 @@ void cv::ocl::device::hog::normalize_hists(int nbins,
openCLExecuteKernel(clCxt, &objdetect_hog, kernelName, globalThreads,
localThreads, args, -1, -1, "-D CPU");
else
{
cl_kernel kernel = openCLGetKernelFromSource(clCxt, &objdetect_hog, kernelName);
int wave_size = queryDeviceInfo<WAVEFRONT_SIZE, int>(kernel);
char opt[32] = {0};
sprintf(opt, "-D WAVE_SIZE=%d", wave_size);
openCLExecuteKernel(clCxt, &objdetect_hog, kernelName, globalThreads,
localThreads, args, -1, -1);
localThreads, args, -1, -1, opt);
}
}
void cv::ocl::device::hog::classify_hists(int win_height, int win_width,
@ -1879,8 +1885,14 @@ void cv::ocl::device::hog::classify_hists(int win_height, int win_width,
openCLExecuteKernel(clCxt, &objdetect_hog, kernelName, globalThreads,
localThreads, args, -1, -1, "-D CPU");
else
{
cl_kernel kernel = openCLGetKernelFromSource(clCxt, &objdetect_hog, kernelName);
int wave_size = queryDeviceInfo<WAVEFRONT_SIZE, int>(kernel);
char opt[32] = {0};
sprintf(opt, "-D WAVE_SIZE=%d", wave_size);
openCLExecuteKernel(clCxt, &objdetect_hog, kernelName, globalThreads,
localThreads, args, -1, -1);
localThreads, args, -1, -1, opt);
}
}
void cv::ocl::device::hog::extract_descrs_by_rows(int win_height, int win_width,

20
modules/ocl/src/opencl/objdetect_hog.cl
View File

@ -133,7 +133,9 @@ __kernel void compute_hists_lut_kernel(
final_hist[(cell_x * 2 + cell_y) * cnbins + bin_id] =
hist_[0] + hist_[1] + hist_[2];
}
#ifdef CPU
barrier(CLK_LOCAL_MEM_FENCE);
#endif
int tid = (cell_y * CELLS_PER_BLOCK_Y + cell_x) * 12 + cell_thread_x;
if ((tid < cblock_hist_size) && (gid < blocks_total))
@ -225,8 +227,9 @@ __kernel void compute_hists_kernel(
final_hist[(cell_x * 2 + cell_y) * cnbins + bin_id] =
hist_[0] + hist_[1] + hist_[2];
}
#ifdef CPU
barrier(CLK_LOCAL_MEM_FENCE);
#endif
int tid = (cell_y * CELLS_PER_BLOCK_Y + cell_x) * 12 + cell_thread_x;
if ((tid < cblock_hist_size) && (gid < blocks_total))
{
@ -318,6 +321,10 @@ float reduce_smem(volatile __local float* smem, int size)
if (tid < 32)
{
if (size >= 64) smem[tid] = sum = sum + smem[tid + 32];
#if WAVE_SIZE < 32
} barrier(CLK_LOCAL_MEM_FENCE);
if (tid < 16) {
#endif
if (size >= 32) smem[tid] = sum = sum + smem[tid + 16];
if (size >= 16) smem[tid] = sum = sum + smem[tid + 8];
if (size >= 8) smem[tid] = sum = sum + smem[tid + 4];
@ -418,6 +425,9 @@ __kernel void classify_hists_180_kernel(
{
smem[tid] = product = product + smem[tid + 32];
}
#if WAVE_SIZE < 32
barrier(CLK_LOCAL_MEM_FENCE);
#endif
if (tid < 16)
{
smem[tid] = product = product + smem[tid + 16];
@ -487,6 +497,10 @@ __kernel void classify_hists_252_kernel(
if (tid < 32)
{
smem[tid] = product = product + smem[tid + 32];
#if WAVE_SIZE < 32
} barrier(CLK_LOCAL_MEM_FENCE);
if (tid < 16) {
#endif
smem[tid] = product = product + smem[tid + 16];
smem[tid] = product = product + smem[tid + 8];
smem[tid] = product = product + smem[tid + 4];
@ -553,6 +567,10 @@ __kernel void classify_hists_kernel(
if (tid < 32)
{
smem[tid] = product = product + smem[tid + 32];
#if WAVE_SIZE < 32
} barrier(CLK_LOCAL_MEM_FENCE);
if (tid < 16) {
#endif
smem[tid] = product = product + smem[tid + 16];
smem[tid] = product = product + smem[tid + 8];
smem[tid] = product = product + smem[tid + 4];

37
modules/ocl/src/opencl/stereobm.cl
View File

@ -258,27 +258,13 @@ float sobel(__global unsigned char *input, int x, int y, int rows, int cols)
float CalcSums(__local float *cols, __local float *cols_cache, int winsz)
{
float cache = 0;
float cache2 = 0;
int winsz2 = winsz/2;
unsigned int cache = cols[0];
int x = get_local_id(0);
int group_size_x = get_local_size(0);
for(int i = 1; i <= winsz2; i++)
#pragma unroll
for(int i = 1; i <= winsz; i++)
cache += cols[i];
cols_cache[0] = cache;
barrier(CLK_LOCAL_MEM_FENCE);
if (x < group_size_x - winsz2)
cache2 = cols_cache[winsz2];
else
for(int i = winsz2 + 1; i < winsz; i++)
cache2 += cols[i];
return cols[0] + cache + cache2;
return cache;
}
#define RpT (2 * ROWSperTHREAD) // got experimentally
@ -301,8 +287,7 @@ __kernel void textureness_kernel(__global unsigned char *disp, int disp_rows, in
int beg_row = group_id_y * RpT;
int end_row = min(beg_row + RpT, disp_rows);
// if (x < disp_cols)
// {
int y = beg_row;
float sum = 0;
@ -340,11 +325,15 @@ __kernel void textureness_kernel(__global unsigned char *disp, int disp_rows, in
}
barrier(CLK_LOCAL_MEM_FENCE);
float sum_win = CalcSums(cols, cols_cache + local_id_x, winsz) * 255;
if (sum_win < threshold)
disp[y * disp_step + x] = 0;
if (x < disp_cols)
{
float sum_win = CalcSums(cols, cols_cache + local_id_x, winsz) * 255;
if (sum_win < threshold)
disp[y * disp_step + x] = 0;
}
barrier(CLK_LOCAL_MEM_FENCE);
}
// }
}

2
modules/ocl/test/main.cpp
View File

@ -118,6 +118,8 @@ int main(int argc, char **argv)
setDevice(oclinfo[pid], device);
setBinaryDiskCache(CACHE_UPDATE);
cout << "Device type:" << type << endl << "Device name:" << oclinfo[pid].DeviceName[device] << endl;
return RUN_ALL_TESTS();
}

6
modules/ocl/test/test_objdetect.cpp
View File

@ -146,17 +146,17 @@ TEST_P(HOG, Detect)
if (winSize.width == 48 && winSize.height == 96)
{
// daimler's base
ocl_hog.setSVMDetector(ocl_hog.getPeopleDetector48x96());
ocl_hog.setSVMDetector(hog.getDaimlerPeopleDetector());
hog.setSVMDetector(hog.getDaimlerPeopleDetector());
}
else if (winSize.width == 64 && winSize.height == 128)
{
ocl_hog.setSVMDetector(ocl_hog.getPeopleDetector64x128());
ocl_hog.setSVMDetector(hog.getDefaultPeopleDetector());
hog.setSVMDetector(hog.getDefaultPeopleDetector());
}
else
{
ocl_hog.setSVMDetector(ocl_hog.getDefaultPeopleDetector());
ocl_hog.setSVMDetector(hog.getDefaultPeopleDetector());
hog.setSVMDetector(hog.getDefaultPeopleDetector());
}

2
samples/ocl/CMakeLists.txt
View File

@ -27,7 +27,7 @@ if(BUILD_EXAMPLES AND OCV_DEPENDENCIES_FOUND)
target_link_libraries(${the_target} ${OPENCV_LINKER_LIBS} ${OPENCV_OCL_SAMPLES_REQUIRED_DEPS})
set_target_properties(${the_target} PROPERTIES
OUTPUT_NAME "${name}_${project}"
OUTPUT_NAME "${project}-example-${name}"
PROJECT_LABEL "(EXAMPLE_${project_upper}) ${name}")
if(ENABLE_SOLUTION_FOLDERS)

108
samples/ocl/clahe.cpp Normal file
View File

@ -0,0 +1,108 @@
#include <iostream>
#include "opencv2/core/core.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/ocl/ocl.hpp"
using namespace cv;
using namespace std;
Ptr<CLAHE> pFilter;
int tilesize;
int cliplimit;
string outfile;
static void TSize_Callback(int pos)
{
if(pos==0)
{
pFilter->setTilesGridSize(Size(1,1));
}
pFilter->setTilesGridSize(Size(tilesize,tilesize));
}
static void Clip_Callback(int)
{
pFilter->setClipLimit(cliplimit);
}
int main(int argc, char** argv)
{
const char* keys =
"{ i | input | | specify input image }"
"{ c | camera | 0 | specify camera id }"
"{ s | use_cpu | false | use cpu algorithm }"
"{ o | output | clahe_output.jpg | specify output save path}";
CommandLineParser cmd(argc, argv, keys);
string infile = cmd.get<string>("i");
outfile = cmd.get<string>("o");
int camid = cmd.get<int>("c");
bool use_cpu = cmd.get<bool>("s");
CvCapture* capture = 0;
bool running = true;
namedWindow("CLAHE");
createTrackbar("Tile Size", "CLAHE", &tilesize, 32, (TrackbarCallback)TSize_Callback);
createTrackbar("Clip Limit", "CLAHE", &cliplimit, 20, (TrackbarCallback)Clip_Callback);
Mat frame, outframe;
ocl::oclMat d_outframe;
int cur_clip;
Size cur_tilesize;
if(use_cpu)
{
pFilter = createCLAHE();
}
else
{
pFilter = ocl::createCLAHE();
}
cur_clip = (int)pFilter->getClipLimit();
cur_tilesize = pFilter->getTilesGridSize();
setTrackbarPos("Tile Size", "CLAHE", cur_tilesize.width);
setTrackbarPos("Clip Limit", "CLAHE", cur_clip);
if(infile != "")
{
frame = imread(infile);
if(frame.empty())
{
cout << "error read image: " << infile << endl;
return -1;
}
}
else
{
capture = cvCaptureFromCAM(camid);
}
cout << "\nControls:\n"
<< "\to - save output image\n"
<< "\tESC - exit\n";
while(running)
{
if(capture)
frame = cvQueryFrame(capture);
else
frame = imread(infile);
if(frame.empty())
{
continue;
}
if(use_cpu)
{
cvtColor(frame, frame, COLOR_BGR2GRAY);
pFilter->apply(frame, outframe);
}
else
{
ocl::oclMat d_frame(frame);
ocl::cvtColor(d_frame, d_outframe, COLOR_BGR2GRAY);
pFilter->apply(d_outframe, d_outframe);
d_outframe.download(outframe);
}
imshow("CLAHE", outframe);
char key = (char)cvWaitKey(3);
if(key == 'o') imwrite(outfile, outframe);
else if(key == 27) running = false;
}
return 0;
}

7
samples/ocl/facedetect.cpp
View File

@ -252,8 +252,13 @@ void Draw(Mat& img, vector<Rect>& faces, double scale)
radius = cvRound((r->width + r->height)*0.25*scale);
circle( img, center, radius, color, 3, 8, 0 );
}
imshow( "result", img );
imwrite( outputName, img );
if(abs(scale-1.0)>.001)
{
resize(img, img, Size((int)(img.cols/scale), (int)(img.rows/scale)));
}
imshow( "result", img );
}

10
samples/ocl/hog.cpp
View File

@ -57,6 +57,7 @@ private:
string vdo_source;
string output;
int camera_id;
bool write_once;
};
int main(int argc, char** argv)
@ -97,6 +98,7 @@ App::App(CommandLineParser& cmd)
<< "\tESC - exit\n"
<< "\tm - change mode GPU <-> CPU\n"
<< "\tg - convert image to gray or not\n"
<< "\to - save output image once, or switch on/off video save\n"
<< "\t1/q - increase/decrease HOG scale\n"
<< "\t2/w - increase/decrease levels count\n"
<< "\t3/e - increase/decrease HOG group threshold\n"
@ -120,6 +122,7 @@ App::App(CommandLineParser& cmd)
hit_threshold = win_width == 48 ? 1.4 : 0.;
scale = 1.05;
gamma_corr = true;
write_once = false;
cout << "Group threshold: " << gr_threshold << endl;
cout << "Levels number: " << nlevels << endl;
@ -254,10 +257,11 @@ void App::run()
workEnd();
if (output!="")
if (output!="" && write_once)
{
if (img_source!="") // wirte image
{
write_once = false;
imwrite(output, img_to_show);
}
else //write video
@ -340,6 +344,10 @@ void App::handleKey(char key)
gamma_corr = !gamma_corr;
cout << "Gamma correction: " << gamma_corr << endl;
break;
case 'o':
case 'O':
write_once = !write_once;
break;
}
}

16
samples/ocl/stereo_match.cpp
View File

@ -49,7 +49,7 @@ struct App
return ss.str();
}
private:
bool running;
bool running, write_once;
Mat left_src, right_src;
Mat left, right;
@ -115,6 +115,7 @@ App::App(CommandLineParser& cmd)
cout << "stereo_match_ocl sample\n";
cout << "\nControls:\n"
<< "\tesc - exit\n"
<< "\to - save output image once\n"
<< "\tp - print current parameters\n"
<< "\tg - convert source images into gray\n"
<< "\tm - change stereo match method\n"
@ -132,6 +133,7 @@ App::App(CommandLineParser& cmd)
else cout << "unknown method!\n";
ndisp = cmd.get<int>("n");
out_img = cmd.get<string>("o");
write_once = false;
}
@ -161,10 +163,8 @@ void App::run()
printParams();
running = true;
bool written = false;
while (running)
{
// Prepare disparity map of specified type
Mat disp;
oclMat d_disp;
@ -192,19 +192,21 @@ void App::run()
csbp(d_left, d_right, d_disp);
break;
}
// Show results
d_disp.download(disp);
workEnd();
if (method != BM)
{
disp.convertTo(disp, 0);
}
putText(disp, text(), Point(5, 25), FONT_HERSHEY_SIMPLEX, 1.0, Scalar::all(255));
imshow("disparity", disp);
if(!written)
if(write_once)
{
imwrite(out_img, disp);
written = true;
write_once = false;
}
handleKey((char)waitKey(3));
}
@ -378,6 +380,10 @@ void App::handleKey(char key)
cout << "level_count: " << csbp.levels << endl;
}
break;
case 'o':
case 'O':
write_once = true;
break;
}
}