mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-25 03:30:34 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add OclCascadeClassifierBuf interface

Browse Source
This commit is contained in:
yao 2013-04-15 17:24:30 +08:00
parent abe2ea59ed
commit 69a0b5dde5
5 changed files with 963 additions and 506 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
modules/ocl/include/opencv2/ocl/ocl.hpp
View File

@ -802,6 +802,44 @@ namespace cv
int minNeighbors, int flags, CvSize minSize = cvSize(0, 0), CvSize maxSize = cvSize(0, 0)); int minNeighbors, int flags, CvSize minSize = cvSize(0, 0), CvSize maxSize = cvSize(0, 0));
}; };
class CV_EXPORTS OclCascadeClassifierBuf : public cv::CascadeClassifier
{
public:
OclCascadeClassifierBuf() :
m_flags(0), initialized(false), m_scaleFactor(0), buffers(NULL) {}
~OclCascadeClassifierBuf() {}
void detectMultiScale(oclMat &image, CV_OUT std::vector<cv::Rect>& faces,
double scaleFactor = 1.1, int minNeighbors = 3, int flags = 0,
Size minSize = Size(), Size maxSize = Size());
void release();
private:
void Init(const int rows, const int cols, double scaleFactor, int flags,
const int outputsz, const size_t localThreads[],
CvSize minSize, CvSize maxSize);
void CreateBaseBufs(const int datasize, const int totalclassifier, const int flags, const int outputsz);
void CreateFactorRelatedBufs(const int rows, const int cols, const int flags,
const double scaleFactor, const size_t localThreads[],
CvSize minSize, CvSize maxSize);
void GenResult(CV_OUT std::vector<cv::Rect>& faces, const std::vector<cv::Rect> &rectList, const std::vector<int> &rweights);
int m_rows;
int m_cols;
int m_flags;
int m_loopcount;
int m_nodenum;
bool findBiggestObject;
bool initialized;
double m_scaleFactor;
Size m_minSize;
Size m_maxSize;
vector<CvSize> sizev;
vector<float> scalev;
oclMat gimg1, gsum, gsqsum;
void * buffers;
};
/////////////////////////////// Pyramid ///////////////////////////////////// /////////////////////////////// Pyramid /////////////////////////////////////

719
modules/ocl/src/haar.cpp
View File

@ -20,6 +20,7 @@
// Jia Haipeng, jiahaipeng95@gmail.com // Jia Haipeng, jiahaipeng95@gmail.com
// Wu Xinglong, wxl370@126.com // Wu Xinglong, wxl370@126.com
// Wang Yao, bitwangyaoyao@gmail.com // Wang Yao, bitwangyaoyao@gmail.com
// Sen Liu, swjtuls1987@126.com
// //
// Redistribution and use in source and binary forms, with or without modification, // Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met: // are permitted provided that the following conditions are met:
@ -842,15 +843,13 @@ static void gpuSetHaarClassifierCascade( CvHaarClassifierCascade *_cascade
} /* j */ } /* j */
} }
} }
CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemStorage *storage, double scaleFactor, CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemStorage *storage, double scaleFactor,
int minNeighbors, int flags, CvSize minSize, CvSize maxSize) int minNeighbors, int flags, CvSize minSize, CvSize maxSize)
{ {
CvHaarClassifierCascade *cascade = oldCascade; CvHaarClassifierCascade *cascade = oldCascade;
//double alltime = (double)cvGetTickCount();
//double t = (double)cvGetTickCount();
const double GROUP_EPS = 0.2; const double GROUP_EPS = 0.2;
oclMat gtemp, gsum1, gtilted1, gsqsum1, gnormImg, gsumcanny;
CvSeq *result_seq = 0; CvSeq *result_seq = 0;
cv::Ptr<CvMemStorage> temp_storage; cv::Ptr<CvMemStorage> temp_storage;
@ -861,7 +860,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
int datasize=0; int datasize=0;
int totalclassifier=0; int totalclassifier=0;
//void *out;
GpuHidHaarClassifierCascade *gcascade; GpuHidHaarClassifierCascade *gcascade;
GpuHidHaarStageClassifier *stage; GpuHidHaarStageClassifier *stage;
GpuHidHaarClassifier *classifier; GpuHidHaarClassifier *classifier;
@ -870,11 +868,8 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
int *candidate; int *candidate;
cl_int status; cl_int status;
// bool doCannyPruning = (flags & CV_HAAR_DO_CANNY_PRUNING) != 0;
bool findBiggestObject = (flags & CV_HAAR_FIND_BIGGEST_OBJECT) != 0; bool findBiggestObject = (flags & CV_HAAR_FIND_BIGGEST_OBJECT) != 0;
// bool roughSearch = (flags & CV_HAAR_DO_ROUGH_SEARCH) != 0;
//double t = 0;
if( maxSize.height == 0 || maxSize.width == 0 ) if( maxSize.height == 0 || maxSize.width == 0 )
{ {
maxSize.height = gimg.rows; maxSize.height = gimg.rows;
@ -896,27 +891,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
if( findBiggestObject ) if( findBiggestObject )
flags &= ~CV_HAAR_SCALE_IMAGE; flags &= ~CV_HAAR_SCALE_IMAGE;
//gtemp = oclMat( gimg.rows, gimg.cols, CV_8UC1);
//gsum1 = oclMat( gimg.rows + 1, gimg.cols + 1, CV_32SC1 );
//gsqsum1 = oclMat( gimg.rows + 1, gimg.cols + 1, CV_32FC1 );
if( !cascade->hid_cascade ) if( !cascade->hid_cascade )
/*out = (void *)*/gpuCreateHidHaarClassifierCascade(cascade, &datasize, &totalclassifier); gpuCreateHidHaarClassifierCascade(cascade, &datasize, &totalclassifier);
if( cascade->hid_cascade->has_tilted_features )
gtilted1 = oclMat( gimg.rows + 1, gimg.cols + 1, CV_32SC1 );
result_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), storage ); result_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), storage );
if( CV_MAT_CN(gimg.type()) > 1 ) if( CV_MAT_CN(gimg.type()) > 1 )
{ {
oclMat gtemp;
cvtColor( gimg, gtemp, CV_BGR2GRAY ); cvtColor( gimg, gtemp, CV_BGR2GRAY );
gimg = gtemp; gimg = gtemp;
} }
if( findBiggestObject ) if( findBiggestObject )
flags &= ~(CV_HAAR_SCALE_IMAGE | CV_HAAR_DO_CANNY_PRUNING); flags &= ~(CV_HAAR_SCALE_IMAGE | CV_HAAR_DO_CANNY_PRUNING);
//t = (double)cvGetTickCount() - t;
//printf( "before if time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
if( gimg.cols < minSize.width || gimg.rows < minSize.height ) if( gimg.cols < minSize.width || gimg.rows < minSize.height )
CV_Error(CV_StsError, "Image too small"); CV_Error(CV_StsError, "Image too small");
@ -924,12 +912,9 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
if( (flags & CV_HAAR_SCALE_IMAGE) ) if( (flags & CV_HAAR_SCALE_IMAGE) )
{ {
CvSize winSize0 = cascade->orig_window_size; CvSize winSize0 = cascade->orig_window_size;
//float scalefactor = 1.1f;
//float factor = 1.f;
int totalheight = 0; int totalheight = 0;
int indexy = 0; int indexy = 0;
CvSize sz; CvSize sz;
//t = (double)cvGetTickCount();
vector<CvSize> sizev; vector<CvSize> sizev;
vector<float> scalev; vector<float> scalev;
for(factor = 1.f;; factor *= scaleFactor) for(factor = 1.f;; factor *= scaleFactor)
@ -950,20 +935,15 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
sizev.push_back(sz); sizev.push_back(sz);
scalev.push_back(factor); scalev.push_back(factor);
} }
//int flag = 0;
oclMat gimg1(gimg.rows, gimg.cols, CV_8UC1); oclMat gimg1(gimg.rows, gimg.cols, CV_8UC1);
oclMat gsum(totalheight + 4, gimg.cols + 1, CV_32SC1); oclMat gsum(totalheight + 4, gimg.cols + 1, CV_32SC1);
oclMat gsqsum(totalheight + 4, gimg.cols + 1, CV_32FC1); oclMat gsqsum(totalheight + 4, gimg.cols + 1, CV_32FC1);
//cl_mem cascadebuffer;
cl_mem stagebuffer; cl_mem stagebuffer;
//cl_mem classifierbuffer;
cl_mem nodebuffer; cl_mem nodebuffer;
cl_mem candidatebuffer; cl_mem candidatebuffer;
cl_mem scaleinfobuffer; cl_mem scaleinfobuffer;
//cl_kernel kernel;
//kernel = openCLGetKernelFromSource(gimg.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade");
cv::Rect roi, roi2; cv::Rect roi, roi2;
cv::Mat imgroi, imgroisq; cv::Mat imgroi, imgroisq;
cv::ocl::oclMat resizeroi, gimgroi, gimgroisq; cv::ocl::oclMat resizeroi, gimgroi, gimgroisq;
@ -971,18 +951,13 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
size_t blocksize = 8; size_t blocksize = 8;
size_t localThreads[3] = { blocksize, blocksize , 1 }; size_t localThreads[3] = { blocksize, blocksize , 1 };
size_t globalThreads[3] = { grp_per_CU *((gsum.clCxt)->computeUnits()) *localThreads[0], size_t globalThreads[3] = { grp_per_CU * gsum.clCxt->computeUnits() *localThreads[0],
localThreads[1], 1 localThreads[1], 1
}; };
int outputsz = 256 * globalThreads[0] / localThreads[0]; int outputsz = 256 * globalThreads[0] / localThreads[0];
int loopcount = sizev.size(); int loopcount = sizev.size();
detect_piramid_info *scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount); detect_piramid_info *scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount);
//t = (double)cvGetTickCount() - t;
// printf( "pre time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
//int *it =scaleinfo;
// t = (double)cvGetTickCount();
for( int i = 0; i < loopcount; i++ ) for( int i = 0; i < loopcount; i++ )
{ {
sz = sizev[i]; sz = sizev[i];
@ -992,7 +967,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
resizeroi = gimg1(roi2); resizeroi = gimg1(roi2);
gimgroi = gsum(roi); gimgroi = gsum(roi);
gimgroisq = gsqsum(roi); gimgroisq = gsqsum(roi);
//scaleinfo[i].rows = gimgroi.rows;
int width = gimgroi.cols - 1 - cascade->orig_window_size.width; int width = gimgroi.cols - 1 - cascade->orig_window_size.width;
int height = gimgroi.rows - 1 - cascade->orig_window_size.height; int height = gimgroi.rows - 1 - cascade->orig_window_size.height;
scaleinfo[i].width_height = (width << 16) | height; scaleinfo[i].width_height = (width << 16) | height;
@ -1000,76 +974,40 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
int grpnumperline = (width + localThreads[0] - 1) / localThreads[0]; int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline; int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
//outputsz +=width*height;
scaleinfo[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp; scaleinfo[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
scaleinfo[i].imgoff = gimgroi.offset >> 2; scaleinfo[i].imgoff = gimgroi.offset >> 2;
scaleinfo[i].factor = factor; scaleinfo[i].factor = factor;
//printf("rows = %d,ystep = %d,width = %d,height = %d,grpnumperline = %d,totalgrp = %d,imgoff = %d,factor = %f\n",
// scaleinfo[i].rows,scaleinfo[i].ystep,scaleinfo[i].width,scaleinfo[i].height,scaleinfo[i].grpnumperline,
// scaleinfo[i].totalgrp,scaleinfo[i].imgoff,scaleinfo[i].factor);
cv::ocl::resize(gimg, resizeroi, Size(sz.width - 1, sz.height - 1), 0, 0, INTER_LINEAR); cv::ocl::resize(gimg, resizeroi, Size(sz.width - 1, sz.height - 1), 0, 0, INTER_LINEAR);
//cv::imwrite("D:\\1.jpg",gimg1);
cv::ocl::integral(resizeroi, gimgroi, gimgroisq); cv::ocl::integral(resizeroi, gimgroi, gimgroisq);
//cv::ocl::oclMat chk(sz.height,sz.width,CV_32SC1),chksq(sz.height,sz.width,CV_32FC1);
//cv::ocl::integral(gimg1, chk, chksq);
//double r = cv::norm(chk,gimgroi,NORM_INF);
//if(r > std::numeric_limits<double>::epsilon())
//{
// printf("failed");
//}
indexy += sz.height; indexy += sz.height;
} }
//int ystep = factor > 2 ? 1 : 2;
// t = (double)cvGetTickCount() - t;
//printf( "resize integral time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
//t = (double)cvGetTickCount();
gcascade = (GpuHidHaarClassifierCascade *)cascade->hid_cascade; gcascade = (GpuHidHaarClassifierCascade *)cascade->hid_cascade;
stage = (GpuHidHaarStageClassifier *)(gcascade + 1); stage = (GpuHidHaarStageClassifier *)(gcascade + 1);
classifier = (GpuHidHaarClassifier *)(stage + gcascade->count); classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
node = (GpuHidHaarTreeNode *)(classifier->node); node = (GpuHidHaarTreeNode *)(classifier->node);
//int m,n;
//m = (gsum.cols - 1 - cascade->orig_window_size.width + ystep - 1)/ystep;
//n = (gsum.rows - 1 - cascade->orig_window_size.height + ystep - 1)/ystep;
//int counter = m*n;
int nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) - int nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) -
sizeof(GpuHidHaarStageClassifier) * gcascade->count - sizeof(GpuHidHaarClassifier) * totalclassifier) / sizeof(GpuHidHaarTreeNode); sizeof(GpuHidHaarStageClassifier) * gcascade->count - sizeof(GpuHidHaarClassifier) * totalclassifier) / sizeof(GpuHidHaarTreeNode);
//if(flag == 0){
candidate = (int *)malloc(4 * sizeof(int) * outputsz);
//memset((char*)candidate,0,4*sizeof(int)*outputsz);
gpuSetImagesForHaarClassifierCascade( cascade,/* &sum1, &sqsum1, _tilted,*/ 1., gsum.step / 4 );
//cascadebuffer = clCreateBuffer(gsum.clCxt->clContext,CL_MEM_READ_ONLY,sizeof(GpuHidHaarClassifierCascade),NULL,&status); candidate = (int *)malloc(4 * sizeof(int) * outputsz);
//openCLVerifyCall(status);
//openCLSafeCall(clEnqueueWriteBuffer(gsum.clCxt->clCmdQueue,cascadebuffer,1,0,sizeof(GpuHidHaarClassifierCascade),gcascade,0,NULL,NULL)); gpuSetImagesForHaarClassifierCascade( cascade, 1., gsum.step / 4 );
stagebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(GpuHidHaarStageClassifier) * gcascade->count); stagebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(GpuHidHaarStageClassifier) * gcascade->count);
//openCLVerifyCall(status); cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL)); openCLSafeCall(clEnqueueWriteBuffer(qu, stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
//classifierbuffer = clCreateBuffer(gsum.clCxt->clContext,CL_MEM_READ_ONLY,sizeof(GpuHidHaarClassifier)*totalclassifier,NULL,&status);
//status = clEnqueueWriteBuffer(gsum.clCxt->clCmdQueue,classifierbuffer,1,0,sizeof(GpuHidHaarClassifier)*totalclassifier,classifier,0,NULL,NULL);
nodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, nodenum * sizeof(GpuHidHaarTreeNode)); nodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, nodenum * sizeof(GpuHidHaarTreeNode));
//openCLVerifyCall(status);
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), nodebuffer, 1, 0, openCLSafeCall(clEnqueueWriteBuffer(qu, nodebuffer, 1, 0, nodenum * sizeof(GpuHidHaarTreeNode),
nodenum * sizeof(GpuHidHaarTreeNode),
node, 0, NULL, NULL)); node, 0, NULL, NULL));
candidatebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_WRITE_ONLY, 4 * sizeof(int) * outputsz); candidatebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_WRITE_ONLY, 4 * sizeof(int) * outputsz);
//openCLVerifyCall(status);
scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount); scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
//openCLVerifyCall(status); openCLSafeCall(clEnqueueWriteBuffer(qu, scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
//flag = 1;
//}
//t = (double)cvGetTickCount() - t;
//printf( "update time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
//size_t globalThreads[3] = { counter+blocksize*blocksize-counter%(blocksize*blocksize),1,1};
//t = (double)cvGetTickCount();
int startstage = 0; int startstage = 0;
int endstage = gcascade->count; int endstage = gcascade->count;
int startnode = 0; int startnode = 0;
@ -1087,11 +1025,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
pq.s[3] = gcascade->pq3; pq.s[3] = gcascade->pq3;
float correction = gcascade->inv_window_area; float correction = gcascade->inv_window_area;
//int grpnumperline = ((m + localThreads[0] - 1) / localThreads[0]);
//int totalgrp = ((n + localThreads[1] - 1) / localThreads[1])*grpnumperline;
// openCLVerifyKernel(gsum.clCxt, kernel, &blocksize, globalThreads, localThreads);
//openCLSafeCall(clSetKernelArg(kernel,argcount++,sizeof(cl_mem),(void*)&cascadebuffer));
vector<pair<size_t, const void *> > args; vector<pair<size_t, const void *> > args;
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&stagebuffer )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&stagebuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&scaleinfobuffer )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&scaleinfobuffer ));
@ -1111,28 +1044,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
args.push_back ( make_pair(sizeof(cl_float) , (void *)&correction )); args.push_back ( make_pair(sizeof(cl_float) , (void *)&correction ));
openCLExecuteKernel(gsum.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade", globalThreads, localThreads, args, -1, -1); openCLExecuteKernel(gsum.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade", globalThreads, localThreads, args, -1, -1);
//t = (double)cvGetTickCount() - t;
//printf( "detection time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
//t = (double)cvGetTickCount();
//openCLSafeCall(clEnqueueReadBuffer(gsum.clCxt->impl->clCmdQueue, candidatebuffer, 1, 0, 4 * sizeof(int)*outputsz, candidate, 0, NULL, NULL));
openCLReadBuffer( gsum.clCxt, candidatebuffer, candidate, 4 * sizeof(int)*outputsz ); openCLReadBuffer( gsum.clCxt, candidatebuffer, candidate, 4 * sizeof(int)*outputsz );
for(int i = 0; i < outputsz; i++) for(int i = 0; i < outputsz; i++)
if(candidate[4 * i + 2] != 0) if(candidate[4 * i + 2] != 0)
allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1], candidate[4 * i + 2], candidate[4 * i + 3])); allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1],
// t = (double)cvGetTickCount() - t; candidate[4 * i + 2], candidate[4 * i + 3]));
//printf( "post time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
//t = (double)cvGetTickCount();
free(scaleinfo); free(scaleinfo);
free(candidate); free(candidate);
//openCLSafeCall(clReleaseMemObject(cascadebuffer));
openCLSafeCall(clReleaseMemObject(stagebuffer)); openCLSafeCall(clReleaseMemObject(stagebuffer));
openCLSafeCall(clReleaseMemObject(scaleinfobuffer)); openCLSafeCall(clReleaseMemObject(scaleinfobuffer));
openCLSafeCall(clReleaseMemObject(nodebuffer)); openCLSafeCall(clReleaseMemObject(nodebuffer));
openCLSafeCall(clReleaseMemObject(candidatebuffer)); openCLSafeCall(clReleaseMemObject(candidatebuffer));
// openCLSafeCall(clReleaseKernel(kernel));
//t = (double)cvGetTickCount() - t;
//printf( "release time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
} }
else else
{ {
@ -1150,7 +1075,6 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
classifier = (GpuHidHaarClassifier *)(stage + gcascade->count); classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
node = (GpuHidHaarTreeNode *)(classifier->node); node = (GpuHidHaarTreeNode *)(classifier->node);
cl_mem stagebuffer; cl_mem stagebuffer;
//cl_mem classifierbuffer;
cl_mem nodebuffer; cl_mem nodebuffer;
cl_mem candidatebuffer; cl_mem candidatebuffer;
cl_mem scaleinfobuffer; cl_mem scaleinfobuffer;
@ -1187,24 +1111,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
size_t blocksize = 8; size_t blocksize = 8;
size_t localThreads[3] = { blocksize, blocksize , 1 }; size_t localThreads[3] = { blocksize, blocksize , 1 };
size_t globalThreads[3] = { grp_per_CU *gsum.clCxt->computeUnits() *localThreads[0], size_t globalThreads[3] = { grp_per_CU *gsum.clCxt->computeUnits() *localThreads[0],
localThreads[1], 1 localThreads[1], 1 };
};
int outputsz = 256 * globalThreads[0] / localThreads[0]; int outputsz = 256 * globalThreads[0] / localThreads[0];
int nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) - int nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) -
sizeof(GpuHidHaarStageClassifier) * gcascade->count - sizeof(GpuHidHaarClassifier) * totalclassifier) / sizeof(GpuHidHaarTreeNode); sizeof(GpuHidHaarStageClassifier) * gcascade->count - sizeof(GpuHidHaarClassifier) * totalclassifier) / sizeof(GpuHidHaarTreeNode);
nodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, nodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY,
nodenum * sizeof(GpuHidHaarTreeNode)); nodenum * sizeof(GpuHidHaarTreeNode));
//openCLVerifyCall(status); cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), nodebuffer, 1, 0, openCLSafeCall(clEnqueueWriteBuffer(qu, nodebuffer, 1, 0,
nodenum * sizeof(GpuHidHaarTreeNode), nodenum * sizeof(GpuHidHaarTreeNode),
node, 0, NULL, NULL)); node, 0, NULL, NULL));
cl_mem newnodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_WRITE, cl_mem newnodebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_WRITE,
loopcount * nodenum * sizeof(GpuHidHaarTreeNode)); loopcount * nodenum * sizeof(GpuHidHaarTreeNode));
int startstage = 0; int startstage = 0;
int endstage = gcascade->count; int endstage = gcascade->count;
//cl_kernel kernel;
//kernel = openCLGetKernelFromSource(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2");
//cl_kernel kernel2 = openCLGetKernelFromSource(gimg.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier");
for(int i = 0; i < loopcount; i++) for(int i = 0; i < loopcount; i++)
{ {
sz = sizev[i]; sz = sizev[i];
@ -1223,7 +1143,7 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
int height = (gsum.rows - 1 - sz.height + ystep - 1) / ystep; int height = (gsum.rows - 1 - sz.height + ystep - 1) / ystep;
int grpnumperline = (width + localThreads[0] - 1) / localThreads[0]; int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline; int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
//outputsz +=width*height;
scaleinfo[i].width_height = (width << 16) | height; scaleinfo[i].width_height = (width << 16) | height;
scaleinfo[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp; scaleinfo[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
scaleinfo[i].imgoff = 0; scaleinfo[i].imgoff = 0;
@ -1241,28 +1161,20 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
size_t globalThreads2[3] = {nodenum, 1, 1}; size_t globalThreads2[3] = {nodenum, 1, 1};
openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier", globalThreads2, NULL/*localThreads2*/, args1, -1, -1); openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier", globalThreads2, NULL/*localThreads2*/, args1, -1, -1);
//clEnqueueNDRangeKernel(gsum.clCxt->impl->clCmdQueue, kernel2, 1, NULL, globalThreads2, 0, 0, NULL, NULL);
//clFinish(gsum.clCxt->impl->clCmdQueue);
} }
//clReleaseKernel(kernel2);
int step = gsum.step / 4; int step = gsum.step / 4;
int startnode = 0; int startnode = 0;
int splitstage = 3; int splitstage = 3;
int splitnode = stage[0].count + stage[1].count + stage[2].count;
stagebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(GpuHidHaarStageClassifier) * gcascade->count); stagebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(GpuHidHaarStageClassifier) * gcascade->count);
//openCLVerifyCall(status); openCLSafeCall(clEnqueueWriteBuffer(qu, stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
candidatebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR, 4 * sizeof(int) * outputsz); candidatebuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR, 4 * sizeof(int) * outputsz);
//openCLVerifyCall(status);
scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount); scaleinfobuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
//openCLVerifyCall(status); openCLSafeCall(clEnqueueWriteBuffer(qu, scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), scaleinfobuffer, 1, 0, sizeof(detect_piramid_info)*loopcount, scaleinfo, 0, NULL, NULL));
pbuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(cl_int4) * loopcount); pbuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(cl_int4) * loopcount);
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), pbuffer, 1, 0, sizeof(cl_int4)*loopcount, p, 0, NULL, NULL)); openCLSafeCall(clEnqueueWriteBuffer(qu, pbuffer, 1, 0, sizeof(cl_int4)*loopcount, p, 0, NULL, NULL));
correctionbuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(cl_float) * loopcount); correctionbuffer = openCLCreateBuffer(gsum.clCxt, CL_MEM_READ_ONLY, sizeof(cl_float) * loopcount);
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), correctionbuffer, 1, 0, sizeof(cl_float)*loopcount, correction, 0, NULL, NULL)); openCLSafeCall(clEnqueueWriteBuffer(qu, correctionbuffer, 1, 0, sizeof(cl_float)*loopcount, correction, 0, NULL, NULL));
//int argcount = 0;
vector<pair<size_t, const void *> > args; vector<pair<size_t, const void *> > args;
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&stagebuffer )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&stagebuffer ));
@ -1271,22 +1183,21 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsum.data )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsum.data ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsqsum.data )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsqsum.data ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&candidatebuffer )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&candidatebuffer ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&gsum.rows ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&gsum.cols ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&step )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&step ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&loopcount )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&loopcount ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&startstage )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&startstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitstage )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&endstage )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&endstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&startnode )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&startnode ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitnode ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&pbuffer )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&pbuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&correctionbuffer )); args.push_back ( make_pair(sizeof(cl_mem) , (void *)&correctionbuffer ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&nodenum )); args.push_back ( make_pair(sizeof(cl_int) , (void *)&nodenum ));
openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2", globalThreads, localThreads, args, -1, -1); openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2", globalThreads, localThreads, args, -1, -1);
//openCLSafeCall(clEnqueueReadBuffer(gsum.clCxt->clCmdQueue,candidatebuffer,1,0,4*sizeof(int)*outputsz,candidate,0,NULL,NULL)); candidate = (int *)clEnqueueMapBuffer(qu, candidatebuffer, 1, CL_MAP_READ, 0, 4 * sizeof(int) * outputsz, 0, 0, 0, &status);
candidate = (int *)clEnqueueMapBuffer((cl_command_queue)gsum.clCxt->oclCommandQueue(), candidatebuffer, 1, CL_MAP_READ, 0, 4 * sizeof(int), 0, 0, 0, &status);
for(int i = 0; i < outputsz; i++) for(int i = 0; i < outputsz; i++)
{ {
@ -1297,7 +1208,7 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
free(scaleinfo); free(scaleinfo);
free(p); free(p);
free(correction); free(correction);
clEnqueueUnmapMemObject((cl_command_queue)gsum.clCxt->oclCommandQueue(), candidatebuffer, candidate, 0, 0, 0); clEnqueueUnmapMemObject(qu, candidatebuffer, candidate, 0, 0, 0);
openCLSafeCall(clReleaseMemObject(stagebuffer)); openCLSafeCall(clReleaseMemObject(stagebuffer));
openCLSafeCall(clReleaseMemObject(scaleinfobuffer)); openCLSafeCall(clReleaseMemObject(scaleinfobuffer));
openCLSafeCall(clReleaseMemObject(nodebuffer)); openCLSafeCall(clReleaseMemObject(nodebuffer));
@ -1306,21 +1217,17 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
openCLSafeCall(clReleaseMemObject(pbuffer)); openCLSafeCall(clReleaseMemObject(pbuffer));
openCLSafeCall(clReleaseMemObject(correctionbuffer)); openCLSafeCall(clReleaseMemObject(correctionbuffer));
} }
//t = (double)cvGetTickCount() ;
cvFree(&cascade->hid_cascade); cvFree(&cascade->hid_cascade);
// printf("%d\n",globalcounter);
rectList.resize(allCandidates.size()); rectList.resize(allCandidates.size());
if(!allCandidates.empty()) if(!allCandidates.empty())
std::copy(allCandidates.begin(), allCandidates.end(), rectList.begin()); std::copy(allCandidates.begin(), allCandidates.end(), rectList.begin());
//cout << "count = " << rectList.size()<< endl;
if( minNeighbors != 0 || findBiggestObject ) if( minNeighbors != 0 || findBiggestObject )
groupRectangles(rectList, rweights, std::max(minNeighbors, 1), GROUP_EPS); groupRectangles(rectList, rweights, std::max(minNeighbors, 1), GROUP_EPS);
else else
rweights.resize(rectList.size(), 0); rweights.resize(rectList.size(), 0);
if( findBiggestObject && rectList.size() ) if( findBiggestObject && rectList.size() )
{ {
CvAvgComp result_comp = {{0, 0, 0, 0}, 0}; CvAvgComp result_comp = {{0, 0, 0, 0}, 0};
@ -1346,13 +1253,565 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
cvSeqPush( result_seq, &c ); cvSeqPush( result_seq, &c );
} }
} }
//t = (double)cvGetTickCount() - t;
//printf( "get face time = %g ms\n", t/((double)cvGetTickFrequency()*1000.) );
//alltime = (double)cvGetTickCount() - alltime;
//printf( "all time = %g ms\n", alltime/((double)cvGetTickFrequency()*1000.) );
return result_seq; return result_seq;
} }
struct OclBuffers
{
cl_mem stagebuffer;
cl_mem nodebuffer;
cl_mem candidatebuffer;
cl_mem scaleinfobuffer;
cl_mem pbuffer;
cl_mem correctionbuffer;
cl_mem newnodebuffer;
};
struct getRect
{
Rect operator()(const CvAvgComp &e) const
{
return e.rect;
}
};
void cv::ocl::OclCascadeClassifierBuf::detectMultiScale(oclMat &gimg, CV_OUT std::vector<cv::Rect>& faces,
double scaleFactor, int minNeighbors, int flags,
Size minSize, Size maxSize)
{
int blocksize = 8;
int grp_per_CU = 12;
size_t localThreads[3] = { blocksize, blocksize, 1 };
size_t globalThreads[3] = { grp_per_CU * Context::getContext()->computeUnits() * localThreads[0],
localThreads[1],
1 };
int outputsz = 256 * globalThreads[0] / localThreads[0];
Init(gimg.rows, gimg.cols, scaleFactor, flags, outputsz, localThreads, minSize, maxSize);
const double GROUP_EPS = 0.2;
cv::ConcurrentRectVector allCandidates;
std::vector<cv::Rect> rectList;
std::vector<int> rweights;
CvHaarClassifierCascade *cascade = oldCascade;
GpuHidHaarClassifierCascade *gcascade;
GpuHidHaarStageClassifier *stage;
GpuHidHaarClassifier *classifier;
GpuHidHaarTreeNode *node;
if( CV_MAT_DEPTH(gimg.type()) != CV_8U )
CV_Error( CV_StsUnsupportedFormat, "Only 8-bit images are supported" );
if( CV_MAT_CN(gimg.type()) > 1 )
{
oclMat gtemp;
cvtColor( gimg, gtemp, CV_BGR2GRAY );
gimg = gtemp;
}
int *candidate;
if( (flags & CV_HAAR_SCALE_IMAGE) )
{
int indexy = 0;
CvSize sz;
cv::Rect roi, roi2;
cv::Mat imgroi, imgroisq;
cv::ocl::oclMat resizeroi, gimgroi, gimgroisq;
for( int i = 0; i < m_loopcount; i++ )
{
sz = sizev[i];
roi = Rect(0, indexy, sz.width, sz.height);
roi2 = Rect(0, 0, sz.width - 1, sz.height - 1);
resizeroi = gimg1(roi2);
gimgroi = gsum(roi);
gimgroisq = gsqsum(roi);
cv::ocl::resize(gimg, resizeroi, Size(sz.width - 1, sz.height - 1), 0, 0, INTER_LINEAR);
cv::ocl::integral(resizeroi, gimgroi, gimgroisq);
indexy += sz.height;
}
gcascade = (GpuHidHaarClassifierCascade *)(cascade->hid_cascade);
stage = (GpuHidHaarStageClassifier *)(gcascade + 1);
classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
node = (GpuHidHaarTreeNode *)(classifier->node);
gpuSetImagesForHaarClassifierCascade( cascade, 1., gsum.step / 4 );
cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->stagebuffer, 1, 0,
sizeof(GpuHidHaarStageClassifier) * gcascade->count,
stage, 0, NULL, NULL));
openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->nodebuffer, 1, 0,
m_nodenum * sizeof(GpuHidHaarTreeNode),
node, 0, NULL, NULL));
int startstage = 0;
int endstage = gcascade->count;
int startnode = 0;
int pixelstep = gsum.step / 4;
int splitstage = 3;
int splitnode = stage[0].count + stage[1].count + stage[2].count;
cl_int4 p, pq;
p.s[0] = gcascade->p0;
p.s[1] = gcascade->p1;
p.s[2] = gcascade->p2;
p.s[3] = gcascade->p3;
pq.s[0] = gcascade->pq0;
pq.s[1] = gcascade->pq1;
pq.s[2] = gcascade->pq2;
pq.s[3] = gcascade->pq3;
float correction = gcascade->inv_window_area;
vector<pair<size_t, const void *> > args;
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->stagebuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->scaleinfobuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->nodebuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsum.data ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsqsum.data ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->candidatebuffer ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&pixelstep ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&m_loopcount ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&startstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&endstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&startnode ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitnode ));
args.push_back ( make_pair(sizeof(cl_int4) , (void *)&p ));
args.push_back ( make_pair(sizeof(cl_int4) , (void *)&pq ));
args.push_back ( make_pair(sizeof(cl_float) , (void *)&correction ));
openCLExecuteKernel(gsum.clCxt, &haarobjectdetect, "gpuRunHaarClassifierCascade", globalThreads, localThreads, args, -1, -1);
candidate = (int *)malloc(4 * sizeof(int) * outputsz);
memset(candidate, 0, 4 * sizeof(int) * outputsz);
openCLReadBuffer( gsum.clCxt, ((OclBuffers *)buffers)->candidatebuffer, candidate, 4 * sizeof(int)*outputsz );
for(int i = 0; i < outputsz; i++)
if(candidate[4 * i + 2] != 0)
allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1],
candidate[4 * i + 2], candidate[4 * i + 3]));
free((void *)candidate);
candidate = NULL;
}
else
{
cv::ocl::integral(gimg, gsum, gsqsum);
gpuSetHaarClassifierCascade(cascade);
gcascade = (GpuHidHaarClassifierCascade *)cascade->hid_cascade;
stage = (GpuHidHaarStageClassifier *)(gcascade + 1);
classifier = (GpuHidHaarClassifier *)(stage + gcascade->count);
node = (GpuHidHaarTreeNode *)(classifier->node);
cl_command_queue qu = (cl_command_queue)gsum.clCxt->oclCommandQueue();
openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->nodebuffer, 1, 0,
m_nodenum * sizeof(GpuHidHaarTreeNode),
node, 0, NULL, NULL));
cl_int4 *p = (cl_int4 *)malloc(sizeof(cl_int4) * m_loopcount);
float *correction = (float *)malloc(sizeof(float) * m_loopcount);
int startstage = 0;
int endstage = gcascade->count;
double factor;
for(int i = 0; i < m_loopcount; i++)
{
factor = scalev[i];
int equRect_x = (int)(factor * gcascade->p0 + 0.5);
int equRect_y = (int)(factor * gcascade->p1 + 0.5);
int equRect_w = (int)(factor * gcascade->p3 + 0.5);
int equRect_h = (int)(factor * gcascade->p2 + 0.5);
p[i].s[0] = equRect_x;
p[i].s[1] = equRect_y;
p[i].s[2] = equRect_x + equRect_w;
p[i].s[3] = equRect_y + equRect_h;
correction[i] = 1. / (equRect_w * equRect_h);
int startnodenum = m_nodenum * i;
float factor2 = (float)factor;
vector<pair<size_t, const void *> > args1;
args1.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->nodebuffer ));
args1.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->newnodebuffer ));
args1.push_back ( make_pair(sizeof(cl_float) , (void *)&factor2 ));
args1.push_back ( make_pair(sizeof(cl_float) , (void *)&correction[i] ));
args1.push_back ( make_pair(sizeof(cl_int) , (void *)&startnodenum ));
size_t globalThreads2[3] = {m_nodenum, 1, 1};
openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuscaleclassifier", globalThreads2, NULL/*localThreads2*/, args1, -1, -1);
}
int step = gsum.step / 4;
int startnode = 0;
int splitstage = 3;
openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->stagebuffer, 1, 0, sizeof(GpuHidHaarStageClassifier)*gcascade->count, stage, 0, NULL, NULL));
openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->pbuffer, 1, 0, sizeof(cl_int4)*m_loopcount, p, 0, NULL, NULL));
openCLSafeCall(clEnqueueWriteBuffer(qu, ((OclBuffers *)buffers)->correctionbuffer, 1, 0, sizeof(cl_float)*m_loopcount, correction, 0, NULL, NULL));
vector<pair<size_t, const void *> > args;
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->stagebuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->scaleinfobuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->newnodebuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsum.data ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&gsqsum.data ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->candidatebuffer ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&gsum.rows ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&gsum.cols ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&step ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&m_loopcount ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&startstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&splitstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&endstage ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&startnode ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->pbuffer ));
args.push_back ( make_pair(sizeof(cl_mem) , (void *)&((OclBuffers *)buffers)->correctionbuffer ));
args.push_back ( make_pair(sizeof(cl_int) , (void *)&m_nodenum ));
openCLExecuteKernel(gsum.clCxt, &haarobjectdetect_scaled2, "gpuRunHaarClassifierCascade_scaled2", globalThreads, localThreads, args, -1, -1);
candidate = (int *)clEnqueueMapBuffer(qu, ((OclBuffers *)buffers)->candidatebuffer, 1, CL_MAP_READ, 0, 4 * sizeof(int) * outputsz, 0, 0, 0, NULL);
for(int i = 0; i < outputsz; i++)
{
if(candidate[4 * i + 2] != 0)
allCandidates.push_back(Rect(candidate[4 * i], candidate[4 * i + 1],
candidate[4 * i + 2], candidate[4 * i + 3]));
}
free(p);
free(correction);
clEnqueueUnmapMemObject(qu, ((OclBuffers *)buffers)->candidatebuffer, candidate, 0, 0, 0);
}
rectList.resize(allCandidates.size());
if(!allCandidates.empty())
std::copy(allCandidates.begin(), allCandidates.end(), rectList.begin());
if( minNeighbors != 0 || findBiggestObject )
groupRectangles(rectList, rweights, std::max(minNeighbors, 1), GROUP_EPS);
else
rweights.resize(rectList.size(), 0);
GenResult(faces, rectList, rweights);
}
void cv::ocl::OclCascadeClassifierBuf::Init(const int rows, const int cols,
double scaleFactor, int flags,
const int outputsz, const size_t localThreads[],
CvSize minSize, CvSize maxSize)
{
CvHaarClassifierCascade *cascade = oldCascade;
if( !CV_IS_HAAR_CLASSIFIER(cascade) )
CV_Error( !cascade ? CV_StsNullPtr : CV_StsBadArg, "Invalid classifier cascade" );
if( scaleFactor <= 1 )
CV_Error( CV_StsOutOfRange, "scale factor must be > 1" );
if( cols < minSize.width || rows < minSize.height )
CV_Error(CV_StsError, "Image too small");
int datasize=0;
int totalclassifier=0;
if( !cascade->hid_cascade )
gpuCreateHidHaarClassifierCascade(cascade, &datasize, &totalclassifier);
if( maxSize.height == 0 || maxSize.width == 0 )
{
maxSize.height = rows;
maxSize.width = cols;
}
findBiggestObject = (flags & CV_HAAR_FIND_BIGGEST_OBJECT) != 0;
if( findBiggestObject )
flags &= ~(CV_HAAR_SCALE_IMAGE | CV_HAAR_DO_CANNY_PRUNING);
CreateBaseBufs(datasize, totalclassifier, flags, outputsz);
CreateFactorRelatedBufs(rows, cols, flags, scaleFactor, localThreads, minSize, maxSize);
m_scaleFactor = scaleFactor;
m_rows = rows;
m_cols = cols;
m_flags = flags;
m_minSize = minSize;
m_maxSize = maxSize;
initialized = true;
}
void cv::ocl::OclCascadeClassifierBuf::CreateBaseBufs(const int datasize, const int totalclassifier,
const int flags, const int outputsz)
{
if (!initialized)
{
buffers = malloc(sizeof(OclBuffers));
size_t tempSize =
sizeof(GpuHidHaarStageClassifier) * ((GpuHidHaarClassifierCascade *)oldCascade->hid_cascade)->count;
m_nodenum = (datasize - sizeof(GpuHidHaarClassifierCascade) - tempSize - sizeof(GpuHidHaarClassifier) * totalclassifier)
/ sizeof(GpuHidHaarTreeNode);
((OclBuffers *)buffers)->stagebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY, tempSize);
((OclBuffers *)buffers)->nodebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY, m_nodenum * sizeof(GpuHidHaarTreeNode));
}
if (initialized
&& ((m_flags & CV_HAAR_SCALE_IMAGE) ^ (flags & CV_HAAR_SCALE_IMAGE)))
{
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->candidatebuffer));
}
if (flags & CV_HAAR_SCALE_IMAGE)
{
((OclBuffers *)buffers)->candidatebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(),
CL_MEM_WRITE_ONLY,
4 * sizeof(int) * outputsz);
}
else
{
((OclBuffers *)buffers)->candidatebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(),
CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR,
4 * sizeof(int) * outputsz);
}
}
void cv::ocl::OclCascadeClassifierBuf::CreateFactorRelatedBufs(
const int rows, const int cols, const int flags,
const double scaleFactor, const size_t localThreads[],
CvSize minSize, CvSize maxSize)
{
if (initialized)
{
if ((m_flags & CV_HAAR_SCALE_IMAGE) && !(flags & CV_HAAR_SCALE_IMAGE))
{
gimg1.release();
gsum.release();
gsqsum.release();
}
else if (!(m_flags & CV_HAAR_SCALE_IMAGE) && (flags & CV_HAAR_SCALE_IMAGE))
{
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->newnodebuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->correctionbuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->pbuffer));
}
else if ((m_flags & CV_HAAR_SCALE_IMAGE) && (flags & CV_HAAR_SCALE_IMAGE))
{
if (fabs(m_scaleFactor - scaleFactor) < 1e-6
&& (rows == m_rows && cols == m_cols)
&& (minSize.width == m_minSize.width)
&& (minSize.height == m_minSize.height)
&& (maxSize.width == m_maxSize.width)
&& (maxSize.height == m_maxSize.height))
{
return;
}
}
else
{
if (fabs(m_scaleFactor - scaleFactor) < 1e-6
&& (rows == m_rows && cols == m_cols)
&& (minSize.width == m_minSize.width)
&& (minSize.height == m_minSize.height)
&& (maxSize.width == m_maxSize.width)
&& (maxSize.height == m_maxSize.height))
{
return;
}
else
{
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->newnodebuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->correctionbuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->pbuffer));
}
}
}
int loopcount;
int indexy = 0;
int totalheight = 0;
double factor;
Rect roi;
CvSize sz;
CvSize winSize0 = oldCascade->orig_window_size;
detect_piramid_info *scaleinfo;
if (flags & CV_HAAR_SCALE_IMAGE)
{
for(factor = 1.f;; factor *= scaleFactor)
{
CvSize winSize = { cvRound(winSize0.width * factor), cvRound(winSize0.height * factor) };
sz.width = cvRound( cols / factor ) + 1;
sz.height = cvRound( rows / factor ) + 1;
CvSize sz1 = { sz.width - winSize0.width - 1, sz.height - winSize0.height - 1 };
if( sz1.width <= 0 || sz1.height <= 0 )
break;
if( winSize.width > maxSize.width || winSize.height > maxSize.height )
break;
if( winSize.width < minSize.width || winSize.height < minSize.height )
continue;
totalheight += sz.height;
sizev.push_back(sz);
scalev.push_back(static_cast<float>(factor));
}
loopcount = sizev.size();
gimg1.create(rows, cols, CV_8UC1);
gsum.create(totalheight + 4, cols + 1, CV_32SC1);
gsqsum.create(totalheight + 4, cols + 1, CV_32FC1);
scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount);
for( int i = 0; i < loopcount; i++ )
{
sz = sizev[i];
roi = Rect(0, indexy, sz.width, sz.height);
int width = sz.width - 1 - oldCascade->orig_window_size.width;
int height = sz.height - 1 - oldCascade->orig_window_size.height;
int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
((detect_piramid_info *)scaleinfo)[i].width_height = (width << 16) | height;
((detect_piramid_info *)scaleinfo)[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
((detect_piramid_info *)scaleinfo)[i].imgoff = gsum(roi).offset >> 2;
((detect_piramid_info *)scaleinfo)[i].factor = scalev[i];
indexy += sz.height;
}
}
else
{
for(factor = 1;
cvRound(factor * winSize0.width) < cols - 10 && cvRound(factor * winSize0.height) < rows - 10;
factor *= scaleFactor)
{
CvSize winSize = { cvRound( winSize0.width * factor ), cvRound( winSize0.height * factor ) };
if( winSize.width < minSize.width || winSize.height < minSize.height )
{
continue;
}
sizev.push_back(winSize);
scalev.push_back(factor);
}
loopcount = scalev.size();
if(loopcount == 0)
{
loopcount = 1;
sizev.push_back(minSize);
scalev.push_back( min(cvRound(minSize.width / winSize0.width), cvRound(minSize.height / winSize0.height)) );
}
((OclBuffers *)buffers)->pbuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY,
sizeof(cl_int4) * loopcount);
((OclBuffers *)buffers)->correctionbuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY,
sizeof(cl_float) * loopcount);
((OclBuffers *)buffers)->newnodebuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_WRITE,
loopcount * m_nodenum * sizeof(GpuHidHaarTreeNode));
scaleinfo = (detect_piramid_info *)malloc(sizeof(detect_piramid_info) * loopcount);
for( int i = 0; i < loopcount; i++ )
{
sz = sizev[i];
factor = scalev[i];
int ystep = cvRound(std::max(2., factor));
int width = (cols - 1 - sz.width + ystep - 1) / ystep;
int height = (rows - 1 - sz.height + ystep - 1) / ystep;
int grpnumperline = (width + localThreads[0] - 1) / localThreads[0];
int totalgrp = ((height + localThreads[1] - 1) / localThreads[1]) * grpnumperline;
((detect_piramid_info *)scaleinfo)[i].width_height = (width << 16) | height;
((detect_piramid_info *)scaleinfo)[i].grpnumperline_totalgrp = (grpnumperline << 16) | totalgrp;
((detect_piramid_info *)scaleinfo)[i].imgoff = 0;
((detect_piramid_info *)scaleinfo)[i].factor = factor;
}
}
if (loopcount != m_loopcount)
{
if (initialized)
{
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->scaleinfobuffer));
}
((OclBuffers *)buffers)->scaleinfobuffer = openCLCreateBuffer(cv::ocl::Context::getContext(), CL_MEM_READ_ONLY, sizeof(detect_piramid_info) * loopcount);
}
openCLSafeCall(clEnqueueWriteBuffer((cl_command_queue)cv::ocl::Context::getContext()->oclCommandQueue(), ((OclBuffers *)buffers)->scaleinfobuffer, 1, 0,
sizeof(detect_piramid_info)*loopcount,
scaleinfo, 0, NULL, NULL));
free(scaleinfo);
m_loopcount = loopcount;
}
void cv::ocl::OclCascadeClassifierBuf::GenResult(CV_OUT std::vector<cv::Rect>& faces,
const std::vector<cv::Rect> &rectList,
const std::vector<int> &rweights)
{
CvSeq *result_seq = cvCreateSeq( 0, sizeof(CvSeq), sizeof(CvAvgComp), cvCreateMemStorage(0) );
if( findBiggestObject && rectList.size() )
{
CvAvgComp result_comp = {{0, 0, 0, 0}, 0};
for( size_t i = 0; i < rectList.size(); i++ )
{
cv::Rect r = rectList[i];
if( r.area() > cv::Rect(result_comp.rect).area() )
{
result_comp.rect = r;
result_comp.neighbors = rweights[i];
}
}
cvSeqPush( result_seq, &result_comp );
}
else
{
for( size_t i = 0; i < rectList.size(); i++ )
{
CvAvgComp c;
c.rect = rectList[i];
c.neighbors = rweights[i];
cvSeqPush( result_seq, &c );
}
}
vector<CvAvgComp> vecAvgComp;
Seq<CvAvgComp>(result_seq).copyTo(vecAvgComp);
faces.resize(vecAvgComp.size());
std::transform(vecAvgComp.begin(), vecAvgComp.end(), faces.begin(), getRect());
}
void cv::ocl::OclCascadeClassifierBuf::release()
{
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->stagebuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->scaleinfobuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->nodebuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->candidatebuffer));
if( (m_flags & CV_HAAR_SCALE_IMAGE) )
{
cvFree(&oldCascade->hid_cascade);
}
else
{
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->newnodebuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->correctionbuffer));
openCLSafeCall(clReleaseMemObject(((OclBuffers *)buffers)->pbuffer));
}
free(buffers);
buffers = NULL;
}
#ifndef _MAX_PATH #ifndef _MAX_PATH
#define _MAX_PATH 1024 #define _MAX_PATH 1024

316
modules/ocl/src/opencl/haarobjectdetect.cl
View File

@ -112,7 +112,7 @@ typedef struct __attribute__((aligned (64))) GpuHidHaarClassifierCascade
} GpuHidHaarClassifierCascade; } GpuHidHaarClassifierCascade;
__kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCascade(//constant GpuHidHaarClassifierCascade * cascade, __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCascade(
global GpuHidHaarStageClassifier * stagecascadeptr, global GpuHidHaarStageClassifier * stagecascadeptr,
global int4 * info, global int4 * info,
global GpuHidHaarTreeNode * nodeptr, global GpuHidHaarTreeNode * nodeptr,
@ -128,12 +128,7 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
const int splitnode, const int splitnode,
const int4 p, const int4 p,
const int4 pq, const int4 pq,
const float correction const float correction)
//const int width,
//const int height,
//const int grpnumperline,
//const int totalgrp
)
{ {
int grpszx = get_local_size(0); int grpszx = get_local_size(0);
int grpszy = get_local_size(1); int grpszy = get_local_size(1);
@ -145,13 +140,8 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int lcl_sz = mul24(grpszx,grpszy); int lcl_sz = mul24(grpszx,grpszy);
int lcl_id = mad24(lclidy,grpszx,lclidx); int lcl_id = mad24(lclidy,grpszx,lclidx);
//assume lcl_sz == 256 or 128 or 64
//int lcl_sz_shift = (lcl_sz == 256) ? 8 : 7;
//lcl_sz_shift = (lcl_sz == 64) ? 6 : lcl_sz_shift;
__local int lclshare[1024]; __local int lclshare[1024];
__local int* lcldata = lclshare;//for save win data
#define OFF 0
__local int* lcldata = lclshare + OFF;//for save win data
__local int* glboutindex = lcldata + 28*28;//for save global out index __local int* glboutindex = lcldata + 28*28;//for save global out index
__local int* lclcount = glboutindex + 1;//for save the numuber of temp pass pixel __local int* lclcount = glboutindex + 1;//for save the numuber of temp pass pixel
__local int* lcloutindex = lclcount + 1;//for save info of temp pass pixel __local int* lcloutindex = lclcount + 1;//for save info of temp pass pixel
@ -181,7 +171,6 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int totalgrp = scaleinfo1.y & 0xffff; int totalgrp = scaleinfo1.y & 0xffff;
int imgoff = scaleinfo1.z; int imgoff = scaleinfo1.z;
float factor = as_float(scaleinfo1.w); float factor = as_float(scaleinfo1.w);
//int ystep =1;// factor > 2.0 ? 1 : 2;
__global const int * sum = sum1 + imgoff; __global const int * sum = sum1 + imgoff;
__global const float * sqsum = sqsum1 + imgoff; __global const float * sqsum = sqsum1 + imgoff;
@ -191,8 +180,6 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int grpidx = grploop - mul24(grpidy, grpnumperline); int grpidx = grploop - mul24(grpidy, grpnumperline);
int x = mad24(grpidx,grpszx,lclidx); int x = mad24(grpidx,grpszx,lclidx);
int y = mad24(grpidy,grpszy,lclidy); int y = mad24(grpidy,grpszy,lclidy);
//candidate_result.x = convert_int_rtn(x*factor);
//candidate_result.y = convert_int_rtn(y*factor);
int grpoffx = x-lclidx; int grpoffx = x-lclidx;
int grpoffy = y-lclidy; int grpoffy = y-lclidy;
@ -211,14 +198,7 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int4 data = *(__global int4*)&sum[glb_off]; int4 data = *(__global int4*)&sum[glb_off];
int lcl_off = mad24(lcl_y, readwidth, lcl_x<<2); int lcl_off = mad24(lcl_y, readwidth, lcl_x<<2);
#if OFF
lcldata[lcl_off] = data.x;
lcldata[lcl_off+1] = data.y;
lcldata[lcl_off+2] = data.z;
lcldata[lcl_off+3] = data.w;
#else
vstore4(data, 0, &lcldata[lcl_off]); vstore4(data, 0, &lcldata[lcl_off]);
#endif
} }
lcloutindex[lcl_id] = 0; lcloutindex[lcl_id] = 0;
@ -231,184 +211,170 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int lcl_off = mad24(lclidy,readwidth,lclidx); int lcl_off = mad24(lclidy,readwidth,lclidx);
int4 cascadeinfo1, cascadeinfo2; int4 cascadeinfo1, cascadeinfo2;
cascadeinfo1 = p; cascadeinfo1 = p;
cascadeinfo2 = pq;// + mad24(y, pixelstep, x); cascadeinfo2 = pq;
cascadeinfo1.x +=lcl_off;
cascadeinfo1.z +=lcl_off;
mean = (lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.x)] - lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.z)] -
lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.x)] + lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.z)])
*correction;
//if((x < width) && (y < height)) int p_offset = mad24(y, pixelstep, x);
cascadeinfo2.x +=p_offset;
cascadeinfo2.z +=p_offset;
variance_norm_factor =sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.x)] - sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.z)] -
sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.x)] + sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.z)];
variance_norm_factor = variance_norm_factor * correction - mean * mean;
variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1.f;
for(int stageloop = start_stage; (stageloop < split_stage) && result; stageloop++ )
{ {
cascadeinfo1.x +=lcl_off; float stage_sum = 0.f;
cascadeinfo1.z +=lcl_off; int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
mean = (lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.x)] - lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.z)] - float stagethreshold = as_float(stageinfo.y);
lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.x)] + lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.z)]) for(int nodeloop = 0; nodeloop < stageinfo.x; nodeloop++ )
*correction; {
__global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter);
int p_offset = mad24(y, pixelstep, x); int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
float nodethreshold = w.w * variance_norm_factor;
cascadeinfo2.x +=p_offset; info1.x +=lcl_off;
cascadeinfo2.z +=p_offset; info1.z +=lcl_off;
variance_norm_factor =sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.x)] - sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.z)] - info2.x +=lcl_off;
sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.x)] + sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.z)]; info2.z +=lcl_off;
variance_norm_factor = variance_norm_factor * correction - mean * mean; float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1.f; lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
//if( cascade->is_stump_based )
//{ classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
for(int stageloop = start_stage; (stageloop < split_stage) && result; stageloop++ ) lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
info3.x +=lcl_off;
info3.z +=lcl_off;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
nodecounter++;
}
result = (stage_sum >= stagethreshold);
}
if(result && (x < width) && (y < height))
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex<<1] = (lclidy << 16) | lclidx;
lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
}
barrier(CLK_LOCAL_MEM_FENCE);
int queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE);
nodecounter = splitnode;
for(int stageloop = split_stage; stageloop< end_stage && queuecount>0; stageloop++)
{
lclcount[0]=0;
barrier(CLK_LOCAL_MEM_FENCE);
int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
float stagethreshold = as_float(stageinfo.y);
int perfscale = queuecount > 4 ? 3 : 2;
int queuecount_loop = (queuecount + (1<<perfscale)-1) >> perfscale;
int lcl_compute_win = lcl_sz >> perfscale;
int lcl_compute_win_id = (lcl_id >>(6-perfscale));
int lcl_loops = (stageinfo.x + lcl_compute_win -1) >> (6-perfscale);
int lcl_compute_id = lcl_id - (lcl_compute_win_id << (6-perfscale));
for(int queueloop=0; queueloop<queuecount_loop; queueloop++)
{ {
float stage_sum = 0.f; float stage_sum = 0.f;
int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop); int temp_coord = lcloutindex[lcl_compute_win_id<<1];
float stagethreshold = as_float(stageinfo.y); float variance_norm_factor = as_float(lcloutindex[(lcl_compute_win_id<<1)+1]);
for(int nodeloop = 0; nodeloop < stageinfo.x; nodeloop++ ) int queue_pixel = mad24(((temp_coord & (int)0xffff0000)>>16),readwidth,temp_coord & 0xffff);
if(lcl_compute_win_id < queuecount)
{ {
__global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter);
int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0])); int tempnodecounter = lcl_compute_id;
int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0])); float part_sum = 0.f;
int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0])); for(int lcl_loop=0; lcl_loop<lcl_loops && tempnodecounter<stageinfo.x; lcl_loop++)
float4 w = *(__global float4*)(&(currentnodeptr->weight[0])); {
float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0])); __global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter + tempnodecounter);
float nodethreshold = w.w * variance_norm_factor;
info1.x +=lcl_off; int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
info1.z +=lcl_off; int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
info2.x +=lcl_off; int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
info2.z +=lcl_off; float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
float nodethreshold = w.w * variance_norm_factor;
float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] - info1.x +=queue_pixel;
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x; info1.z +=queue_pixel;
info2.x +=queue_pixel;
info2.z +=queue_pixel;
float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] - classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y; lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
info3.x +=queue_pixel;
info3.z +=queue_pixel;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
//if((info3.z - info3.x) && (!stageinfo.z)) part_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
//{ tempnodecounter +=lcl_compute_win;
info3.x +=lcl_off; }//end for(int lcl_loop=0;lcl_loop<lcl_loops;lcl_loop++)
info3.z +=lcl_off; partialsum[lcl_id]=part_sum;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
//}
stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
nodecounter++;
} }
result = (stage_sum >= stagethreshold);
}
if(result && (x < width) && (y < height))
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex<<1] = (lclidy << 16) | lclidx;
lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
}
barrier(CLK_LOCAL_MEM_FENCE);
int queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE);
nodecounter = splitnode;
for(int stageloop = split_stage; stageloop< end_stage && queuecount>0; stageloop++)
{
//barrier(CLK_LOCAL_MEM_FENCE);
//if(lcl_id == 0)
lclcount[0]=0;
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount)
int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
float stagethreshold = as_float(stageinfo.y);
int perfscale = queuecount > 4 ? 3 : 2;
int queuecount_loop = (queuecount + (1<<perfscale)-1) >> perfscale;
int lcl_compute_win = lcl_sz >> perfscale;
int lcl_compute_win_id = (lcl_id >>(6-perfscale));
int lcl_loops = (stageinfo.x + lcl_compute_win -1) >> (6-perfscale);
int lcl_compute_id = lcl_id - (lcl_compute_win_id << (6-perfscale));
for(int queueloop=0; queueloop<queuecount_loop/* && lcl_compute_win_id < queuecount*/; queueloop++)
{ {
float stage_sum = 0.f; for(int i=0; i<lcl_compute_win && (lcl_compute_id==0); i++)
int temp_coord = lcloutindex[lcl_compute_win_id<<1];
float variance_norm_factor = as_float(lcloutindex[(lcl_compute_win_id<<1)+1]);
int queue_pixel = mad24(((temp_coord & (int)0xffff0000)>>16),readwidth,temp_coord & 0xffff);
//barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount)
{ {
stage_sum += partialsum[lcl_id+i];
int tempnodecounter = lcl_compute_id;
float part_sum = 0.f;
for(int lcl_loop=0; lcl_loop<lcl_loops && tempnodecounter<stageinfo.x; lcl_loop++)
{
__global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter + tempnodecounter);
int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
float nodethreshold = w.w * variance_norm_factor;
info1.x +=queue_pixel;
info1.z +=queue_pixel;
info2.x +=queue_pixel;
info2.z +=queue_pixel;
float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
//if((info3.z - info3.x) && (!stageinfo.z))
//{
info3.x +=queue_pixel;
info3.z +=queue_pixel;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
//}
part_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
tempnodecounter +=lcl_compute_win;
}//end for(int lcl_loop=0;lcl_loop<lcl_loops;lcl_loop++)
partialsum[lcl_id]=part_sum;
} }
barrier(CLK_LOCAL_MEM_FENCE); if(stage_sum >= stagethreshold && (lcl_compute_id==0))
if(lcl_compute_win_id < queuecount)
{ {
for(int i=0; i<lcl_compute_win && (lcl_compute_id==0); i++) int queueindex = atomic_inc(lclcount);
{ lcloutindex[queueindex<<1] = temp_coord;
stage_sum += partialsum[lcl_id+i]; lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
}
if(stage_sum >= stagethreshold && (lcl_compute_id==0))
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex<<1] = temp_coord;
lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
}
lcl_compute_win_id +=(1<<perfscale);
} }
barrier(CLK_LOCAL_MEM_FENCE); lcl_compute_win_id +=(1<<perfscale);
}//end for(int queueloop=0;queueloop<queuecount_loop;queueloop++) }
//barrier(CLK_LOCAL_MEM_FENCE);
queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
nodecounter += stageinfo.x; }//end for(int queueloop=0;queueloop<queuecount_loop;queueloop++)
}//end for(int stageloop = splitstage; stageloop< endstage && queuecount>0;stageloop++)
//barrier(CLK_LOCAL_MEM_FENCE); queuecount = lclcount[0];
if(lcl_id<queuecount)
{
int temp = lcloutindex[lcl_id<<1];
int x = mad24(grpidx,grpszx,temp & 0xffff);
int y = mad24(grpidy,grpszy,((temp & (int)0xffff0000) >> 16));
temp = glboutindex[0];
int4 candidate_result;
candidate_result.zw = (int2)convert_int_rtn(factor*20.f);
candidate_result.x = convert_int_rtn(x*factor);
candidate_result.y = convert_int_rtn(y*factor);
atomic_inc(glboutindex);
candidate[outputoff+temp+lcl_id] = candidate_result;
}
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
}//end if((x < width) && (y < height)) nodecounter += stageinfo.x;
}//end for(int stageloop = splitstage; stageloop< endstage && queuecount>0;stageloop++)
if(lcl_id<queuecount)
{
int temp = lcloutindex[lcl_id<<1];
int x = mad24(grpidx,grpszx,temp & 0xffff);
int y = mad24(grpidy,grpszy,((temp & (int)0xffff0000) >> 16));
temp = glboutindex[0];
int4 candidate_result;
candidate_result.zw = (int2)convert_int_rtn(factor*20.f);
candidate_result.x = convert_int_rtn(x*factor);
candidate_result.y = convert_int_rtn(y*factor);
atomic_inc(glboutindex);
candidate[outputoff+temp+lcl_id] = candidate_result;
}
barrier(CLK_LOCAL_MEM_FENCE);
}//end for(int grploop=grpidx;grploop<totalgrp;grploop+=grpnumx) }//end for(int grploop=grpidx;grploop<totalgrp;grploop+=grpnumx)
//outputoff +=mul24(width,height);
}//end for(int scalei = 0; scalei <loopcount; scalei++) }//end for(int scalei = 0; scalei <loopcount; scalei++)
} }

294
modules/ocl/src/opencl/haarobjectdetect_scaled2.cl
View File

@ -16,6 +16,7 @@
// //
// @Authors // @Authors
// Wu Xinglong, wxl370@126.com // Wu Xinglong, wxl370@126.com
// Sen Liu, swjtuls1987@126.com
// //
// Redistribution and use in source and binary forms, with or without modification, // Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met: // are permitted provided that the following conditions are met:
@ -52,11 +53,11 @@ typedef struct __attribute__((aligned(128))) GpuHidHaarFeature
{ {
struct __attribute__((aligned(32))) struct __attribute__((aligned(32)))
{ {
int p0 __attribute__((aligned(4))); int p0 __attribute__((aligned(4)));
int p1 __attribute__((aligned(4))); int p1 __attribute__((aligned(4)));
int p2 __attribute__((aligned(4))); int p2 __attribute__((aligned(4)));
int p3 __attribute__((aligned(4))); int p3 __attribute__((aligned(4)));
float weight __attribute__((aligned(4))); float weight __attribute__((aligned(4)));
} }
rect[CV_HAAR_FEATURE_MAX] __attribute__((aligned(32))); rect[CV_HAAR_FEATURE_MAX] __attribute__((aligned(32)));
} }
@ -113,173 +114,168 @@ __kernel void gpuRunHaarClassifierCascade_scaled2(
global const int *restrict sum, global const int *restrict sum,
global const float *restrict sqsum, global const float *restrict sqsum,
global int4 *candidate, global int4 *candidate,
const int rows,
const int cols,
const int step, const int step,
const int loopcount, const int loopcount,
const int start_stage, const int start_stage,
const int split_stage, const int split_stage,
const int end_stage, const int end_stage,
const int startnode, const int startnode,
const int splitnode,
global int4 *p, global int4 *p,
//const int4 * pq,
global float *correction, global float *correction,
const int nodecount) const int nodecount)
{ {
int grpszx = get_local_size(0); int grpszx = get_local_size(0);
int grpszy = get_local_size(1); int grpszy = get_local_size(1);
int grpnumx = get_num_groups(0); int grpnumx = get_num_groups(0);
int grpidx = get_group_id(0); int grpidx = get_group_id(0);
int lclidx = get_local_id(0); int lclidx = get_local_id(0);
int lclidy = get_local_id(1); int lclidy = get_local_id(1);
int lcl_sz = mul24(grpszx, grpszy); int lcl_sz = mul24(grpszx, grpszy);
int lcl_id = mad24(lclidy, grpszx, lclidx); int lcl_id = mad24(lclidy, grpszx, lclidx);
__local int lclshare[1024]; __local int glboutindex[1];
__local int *glboutindex = lclshare + 0; __local int lclcount[1];
__local int *lclcount = glboutindex + 1; __local int lcloutindex[64];
__local int *lcloutindex = lclcount + 1; glboutindex[0] = 0;
__local float *partialsum = (__local float *)(lcloutindex + (lcl_sz << 1)); int outputoff = mul24(grpidx, 256);
glboutindex[0] = 0; candidate[outputoff + (lcl_id << 2)] = (int4)0;
int outputoff = mul24(grpidx, 256); candidate[outputoff + (lcl_id << 2) + 1] = (int4)0;
candidate[outputoff + (lcl_id << 2)] = (int4)0; candidate[outputoff + (lcl_id << 2) + 2] = (int4)0;
candidate[outputoff + (lcl_id << 2) + 1] = (int4)0; candidate[outputoff + (lcl_id << 2) + 3] = (int4)0;
candidate[outputoff + (lcl_id << 2) + 2] = (int4)0; int max_idx = rows * cols - 1;
candidate[outputoff + (lcl_id << 2) + 3] = (int4)0; for (int scalei = 0; scalei < loopcount; scalei++)
{
int4 scaleinfo1;
scaleinfo1 = info[scalei];
int width = (scaleinfo1.x & 0xffff0000) >> 16;
int height = scaleinfo1.x & 0xffff;
int grpnumperline = (scaleinfo1.y & 0xffff0000) >> 16;
int totalgrp = scaleinfo1.y & 0xffff;
float factor = as_float(scaleinfo1.w);
float correction_t = correction[scalei];
int ystep = (int)(max(2.0f, factor) + 0.5f);
for (int scalei = 0; scalei < loopcount; scalei++) for (int grploop = get_group_id(0); grploop < totalgrp; grploop += grpnumx)
{ {
int4 scaleinfo1; int4 cascadeinfo = p[scalei];
scaleinfo1 = info[scalei]; int grpidy = grploop / grpnumperline;
int width = (scaleinfo1.x & 0xffff0000) >> 16; int grpidx = grploop - mul24(grpidy, grpnumperline);
int height = scaleinfo1.x & 0xffff; int ix = mad24(grpidx, grpszx, lclidx);
int grpnumperline = (scaleinfo1.y & 0xffff0000) >> 16; int iy = mad24(grpidy, grpszy, lclidy);
int totalgrp = scaleinfo1.y & 0xffff; int x = ix * ystep;
float factor = as_float(scaleinfo1.w); int y = iy * ystep;
float correction_t = correction[scalei]; lcloutindex[lcl_id] = 0;
int ystep = (int)(max(2.0f, factor) + 0.5f); lclcount[0] = 0;
int nodecounter;
float mean, variance_norm_factor;
//if((ix < width) && (iy < height))
{
const int p_offset = mad24(y, step, x);
cascadeinfo.x += p_offset;
cascadeinfo.z += p_offset;
mean = (sum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.x), 0, max_idx)] - sum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.z), 0, max_idx)] -
sum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.x), 0, max_idx)] + sum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.z), 0, max_idx)])
* correction_t;
variance_norm_factor = sqsum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.x), 0, max_idx)] - sqsum[clamp(mad24(cascadeinfo.y, step, cascadeinfo.z), 0, max_idx)] -
sqsum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.x), 0, max_idx)] + sqsum[clamp(mad24(cascadeinfo.w, step, cascadeinfo.z), 0, max_idx)];
variance_norm_factor = variance_norm_factor * correction_t - mean * mean;
variance_norm_factor = variance_norm_factor >= 0.f ? sqrt(variance_norm_factor) : 1.f;
bool result = true;
nodecounter = startnode + nodecount * scalei;
for (int grploop = get_group_id(0); grploop < totalgrp; grploop += grpnumx) for (int stageloop = start_stage; (stageloop < end_stage) && result; stageloop++)
{ {
int4 cascadeinfo = p[scalei]; float stage_sum = 0.f;
int grpidy = grploop / grpnumperline; int stagecount = stagecascadeptr[stageloop].count;
int grpidx = grploop - mul24(grpidy, grpnumperline); for (int nodeloop = 0; nodeloop < stagecount; nodeloop++)
int ix = mad24(grpidx, grpszx, lclidx); {
int iy = mad24(grpidy, grpszy, lclidy); __global GpuHidHaarTreeNode *currentnodeptr = (nodeptr + nodecounter);
int x = ix * ystep; int4 info1 = *(__global int4 *)(&(currentnodeptr->p[0][0]));
int y = iy * ystep; int4 info2 = *(__global int4 *)(&(currentnodeptr->p[1][0]));
lcloutindex[lcl_id] = 0; int4 info3 = *(__global int4 *)(&(currentnodeptr->p[2][0]));
lclcount[0] = 0; float4 w = *(__global float4 *)(&(currentnodeptr->weight[0]));
int result = 1, nodecounter; float2 alpha2 = *(__global float2 *)(&(currentnodeptr->alpha[0]));
float mean, variance_norm_factor; float nodethreshold = w.w * variance_norm_factor;
//if((ix < width) && (iy < height)) info1.x += p_offset;
{ info1.z += p_offset;
const int p_offset = mad24(y, step, x); info2.x += p_offset;
cascadeinfo.x += p_offset; info2.z += p_offset;
cascadeinfo.z += p_offset; float classsum = (sum[clamp(mad24(info1.y, step, info1.x), 0, max_idx)] - sum[clamp(mad24(info1.y, step, info1.z), 0, max_idx)] -
mean = (sum[mad24(cascadeinfo.y, step, cascadeinfo.x)] - sum[mad24(cascadeinfo.y, step, cascadeinfo.z)] - sum[clamp(mad24(info1.w, step, info1.x), 0, max_idx)] + sum[clamp(mad24(info1.w, step, info1.z), 0, max_idx)]) * w.x;
sum[mad24(cascadeinfo.w, step, cascadeinfo.x)] + sum[mad24(cascadeinfo.w, step, cascadeinfo.z)]) classsum += (sum[clamp(mad24(info2.y, step, info2.x), 0, max_idx)] - sum[clamp(mad24(info2.y, step, info2.z), 0, max_idx)] -
* correction_t; sum[clamp(mad24(info2.w, step, info2.x), 0, max_idx)] + sum[clamp(mad24(info2.w, step, info2.z), 0, max_idx)]) * w.y;
variance_norm_factor = sqsum[mad24(cascadeinfo.y, step, cascadeinfo.x)] - sqsum[mad24(cascadeinfo.y, step, cascadeinfo.z)] - info3.x += p_offset;
sqsum[mad24(cascadeinfo.w, step, cascadeinfo.x)] + sqsum[mad24(cascadeinfo.w, step, cascadeinfo.z)]; info3.z += p_offset;
variance_norm_factor = variance_norm_factor * correction_t - mean * mean; classsum += (sum[clamp(mad24(info3.y, step, info3.x), 0, max_idx)] - sum[clamp(mad24(info3.y, step, info3.z), 0, max_idx)] -
variance_norm_factor = variance_norm_factor >= 0.f ? sqrt(variance_norm_factor) : 1.f; sum[clamp(mad24(info3.w, step, info3.x), 0, max_idx)] + sum[clamp(mad24(info3.w, step, info3.z), 0, max_idx)]) * w.z;
result = 1; stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
nodecounter = startnode + nodecount * scalei; nodecounter++;
}
for (int stageloop = start_stage; stageloop < end_stage && result; stageloop++) result = (bool)(stage_sum >= stagecascadeptr[stageloop].threshold);
{
float stage_sum = 0.f;
int4 stageinfo = *(global int4 *)(stagecascadeptr + stageloop);
float stagethreshold = as_float(stageinfo.y);
for (int nodeloop = 0; nodeloop < stageinfo.x; nodeloop++)
{
__global GpuHidHaarTreeNode *currentnodeptr = (nodeptr + nodecounter);
int4 info1 = *(__global int4 *)(&(currentnodeptr->p[0][0]));
int4 info2 = *(__global int4 *)(&(currentnodeptr->p[1][0]));
int4 info3 = *(__global int4 *)(&(currentnodeptr->p[2][0]));
float4 w = *(__global float4 *)(&(currentnodeptr->weight[0]));
float2 alpha2 = *(__global float2 *)(&(currentnodeptr->alpha[0]));
float nodethreshold = w.w * variance_norm_factor;
info1.x += p_offset;
info1.z += p_offset;
info2.x += p_offset;
info2.z += p_offset;
float classsum = (sum[mad24(info1.y, step, info1.x)] - sum[mad24(info1.y, step, info1.z)] -
sum[mad24(info1.w, step, info1.x)] + sum[mad24(info1.w, step, info1.z)]) * w.x;
classsum += (sum[mad24(info2.y, step, info2.x)] - sum[mad24(info2.y, step, info2.z)] -
sum[mad24(info2.w, step, info2.x)] + sum[mad24(info2.w, step, info2.z)]) * w.y;
info3.x += p_offset;
info3.z += p_offset;
classsum += (sum[mad24(info3.y, step, info3.x)] - sum[mad24(info3.y, step, info3.z)] -
sum[mad24(info3.w, step, info3.x)] + sum[mad24(info3.w, step, info3.z)]) * w.z;
stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
nodecounter++;
}
result = (stage_sum >= stagethreshold);
}
if (result && (ix < width) && (iy < height))
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex << 1] = (y << 16) | x;
lcloutindex[(queueindex << 1) + 1] = as_int(variance_norm_factor);
}
barrier(CLK_LOCAL_MEM_FENCE);
int queuecount = lclcount[0];
nodecounter = splitnode + nodecount * scalei;
if (lcl_id < queuecount)
{
int temp = lcloutindex[lcl_id << 1];
int x = temp & 0xffff;
int y = (temp & (int)0xffff0000) >> 16;
temp = glboutindex[0];
int4 candidate_result;
candidate_result.zw = (int2)convert_int_rtn(factor * 20.f);
candidate_result.x = x;
candidate_result.y = y;
atomic_inc(glboutindex);
candidate[outputoff + temp + lcl_id] = candidate_result;
}
barrier(CLK_LOCAL_MEM_FENCE);
}
} }
barrier(CLK_LOCAL_MEM_FENCE);
if (result && (ix < width) && (iy < height))
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex] = (y << 16) | x;
}
barrier(CLK_LOCAL_MEM_FENCE);
int queuecount = lclcount[0];
if (lcl_id < queuecount)
{
int temp = lcloutindex[lcl_id];
int x = temp & 0xffff;
int y = (temp & (int)0xffff0000) >> 16;
temp = atomic_inc(glboutindex);
int4 candidate_result;
candidate_result.zw = (int2)convert_int_rtn(factor * 20.f);
candidate_result.x = x;
candidate_result.y = y;
candidate[outputoff + temp + lcl_id] = candidate_result;
}
barrier(CLK_LOCAL_MEM_FENCE);
}
} }
}
} }
__kernel void gpuscaleclassifier(global GpuHidHaarTreeNode *orinode, global GpuHidHaarTreeNode *newnode, float scale, float weight_scale, int nodenum) __kernel void gpuscaleclassifier(global GpuHidHaarTreeNode *orinode, global GpuHidHaarTreeNode *newnode, float scale, float weight_scale, int nodenum)
{ {
int counter = get_global_id(0); int counter = get_global_id(0);
int tr_x[3], tr_y[3], tr_h[3], tr_w[3], i = 0; int tr_x[3], tr_y[3], tr_h[3], tr_w[3], i = 0;
GpuHidHaarTreeNode t1 = *(orinode + counter); GpuHidHaarTreeNode t1 = *(orinode + counter);
#pragma unroll #pragma unroll
for (i = 0; i < 3; i++) for (i = 0; i < 3; i++)
{ {
tr_x[i] = (int)(t1.p[i][0] * scale + 0.5f); tr_x[i] = (int)(t1.p[i][0] * scale + 0.5f);
tr_y[i] = (int)(t1.p[i][1] * scale + 0.5f); tr_y[i] = (int)(t1.p[i][1] * scale + 0.5f);
tr_w[i] = (int)(t1.p[i][2] * scale + 0.5f); tr_w[i] = (int)(t1.p[i][2] * scale + 0.5f);
tr_h[i] = (int)(t1.p[i][3] * scale + 0.5f); tr_h[i] = (int)(t1.p[i][3] * scale + 0.5f);
} }
t1.weight[0] = t1.p[2][0] ? -(t1.weight[1] * tr_h[1] * tr_w[1] + t1.weight[2] * tr_h[2] * tr_w[2]) / (tr_h[0] * tr_w[0]) : -t1.weight[1] * tr_h[1] * tr_w[1] / (tr_h[0] * tr_w[0]); t1.weight[0] = t1.p[2][0] ? -(t1.weight[1] * tr_h[1] * tr_w[1] + t1.weight[2] * tr_h[2] * tr_w[2]) / (tr_h[0] * tr_w[0]) : -t1.weight[1] * tr_h[1] * tr_w[1] / (tr_h[0] * tr_w[0]);
counter += nodenum; counter += nodenum;
#pragma unroll #pragma unroll
for (i = 0; i < 3; i++) for (i = 0; i < 3; i++)
{ {
newnode[counter].p[i][0] = tr_x[i]; newnode[counter].p[i][0] = tr_x[i];
newnode[counter].p[i][1] = tr_y[i]; newnode[counter].p[i][1] = tr_y[i];
newnode[counter].p[i][2] = tr_x[i] + tr_w[i]; newnode[counter].p[i][2] = tr_x[i] + tr_w[i];
newnode[counter].p[i][3] = tr_y[i] + tr_h[i]; newnode[counter].p[i][3] = tr_y[i] + tr_h[i];
newnode[counter].weight[i] = t1.weight[i] * weight_scale; newnode[counter].weight[i] = t1.weight[i] * weight_scale;
} }
newnode[counter].left = t1.left; newnode[counter].left = t1.left;
newnode[counter].right = t1.right; newnode[counter].right = t1.right;
newnode[counter].threshold = t1.threshold; newnode[counter].threshold = t1.threshold;
newnode[counter].alpha[0] = t1.alpha[0]; newnode[counter].alpha[0] = t1.alpha[0];
newnode[counter].alpha[1] = t1.alpha[1]; newnode[counter].alpha[1] = t1.alpha[1];
} }

102
modules/ocl/test/test_haar.cpp
View File

@ -16,6 +16,7 @@
// //
// @Authors // @Authors
// Jia Haipeng, jiahaipeng95@gmail.com // Jia Haipeng, jiahaipeng95@gmail.com
// Sen Liu, swjutls1987@126.com
// //
// Redistribution and use in source and binary forms, with or without modification, // Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met: // are permitted provided that the following conditions are met:
@ -61,40 +62,31 @@ struct getRect
} }
}; };
PARAM_TEST_CASE(HaarTestBase, int, int) PARAM_TEST_CASE(Haar, double, int)
{ {
//std::vector<cv::ocl::Info> oclinfo;
cv::ocl::OclCascadeClassifier cascade, nestedCascade; cv::ocl::OclCascadeClassifier cascade, nestedCascade;
cv::ocl::OclCascadeClassifierBuf cascadebuf;
cv::CascadeClassifier cpucascade, cpunestedCascade; cv::CascadeClassifier cpucascade, cpunestedCascade;
// Mat img;
double scale; double scale;
int index; int flags;
virtual void SetUp() virtual void SetUp()
{ {
scale = 1.0; scale = GET_PARAM(0);
index = 0; flags = GET_PARAM(1);
string cascadeName = workdir + "../../data/haarcascades/haarcascade_frontalface_alt.xml"; string cascadeName = workdir + "../../data/haarcascades/haarcascade_frontalface_alt.xml";
if( (!cascade.load( cascadeName )) || (!cpucascade.load(cascadeName))) if( (!cascade.load( cascadeName )) || (!cpucascade.load(cascadeName)) || (!cascadebuf.load( cascadeName )))
{ {
cout << "ERROR: Could not load classifier cascade" << endl; cout << "ERROR: Could not load classifier cascade" << endl;
return; return;
} }
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums>0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//cv::ocl::setBinpath("E:\\");
} }
}; };
////////////////////////////////faceDetect///////////////////////////////////////////////// ////////////////////////////////faceDetect/////////////////////////////////////////////////
TEST_P(Haar, FaceDetect)
struct Haar : HaarTestBase {};
TEST_F(Haar, FaceDetect)
{ {
string imgName = workdir + "lena.jpg"; string imgName = workdir + "lena.jpg";
Mat img = imread( imgName, 1 ); Mat img = imread( imgName, 1 );
@ -105,59 +97,65 @@ TEST_F(Haar, FaceDetect)
return ; return ;
} }
//int i = 0;
//double t = 0;
vector<Rect> faces, oclfaces; vector<Rect> faces, oclfaces;
// const static Scalar colors[] = { CV_RGB(0, 0, 255),
// CV_RGB(0, 128, 255),
// CV_RGB(0, 255, 255),
// CV_RGB(0, 255, 0),
// CV_RGB(255, 128, 0),
// CV_RGB(255, 255, 0),
// CV_RGB(255, 0, 0),
// CV_RGB(255, 0, 255)
// } ;
Mat gray, smallImg(cvRound (img.rows / scale), cvRound(img.cols / scale), CV_8UC1 ); Mat gray, smallImg(cvRound (img.rows / scale), cvRound(img.cols / scale), CV_8UC1 );
MemStorage storage(cvCreateMemStorage(0)); MemStorage storage(cvCreateMemStorage(0));
cvtColor( img, gray, CV_BGR2GRAY ); cvtColor( img, gray, CV_BGR2GRAY );
resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR ); resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
equalizeHist( smallImg, smallImg ); equalizeHist( smallImg, smallImg );
cv::ocl::oclMat image; cv::ocl::oclMat image;
CvSeq *_objects; CvSeq *_objects;
image.upload(smallImg); image.upload(smallImg);
_objects = cascade.oclHaarDetectObjects( image, storage, 1.1, _objects = cascade.oclHaarDetectObjects( image, storage, 1.1,
3, 0 3, flags, Size(30, 30), Size(0, 0) );
| CV_HAAR_SCALE_IMAGE
, Size(30, 30), Size(0, 0) );
vector<CvAvgComp> vecAvgComp; vector<CvAvgComp> vecAvgComp;
Seq<CvAvgComp>(_objects).copyTo(vecAvgComp); Seq<CvAvgComp>(_objects).copyTo(vecAvgComp);
oclfaces.resize(vecAvgComp.size()); oclfaces.resize(vecAvgComp.size());
std::transform(vecAvgComp.begin(), vecAvgComp.end(), oclfaces.begin(), getRect()); std::transform(vecAvgComp.begin(), vecAvgComp.end(), oclfaces.begin(), getRect());
cpucascade.detectMultiScale( smallImg, faces, 1.1, cpucascade.detectMultiScale( smallImg, faces, 1.1, 3,
3, 0 flags,
| CV_HAAR_SCALE_IMAGE Size(30, 30), Size(0, 0) );
, Size(30, 30), Size(0, 0) );
EXPECT_EQ(faces.size(), oclfaces.size()); EXPECT_EQ(faces.size(), oclfaces.size());
/* for( vector<Rect>::const_iterator r = faces.begin(); r != faces.end(); r++, i++ )
{
Mat smallImgROI;
Point center;
Scalar color = colors[i%8];
int radius;
center.x = cvRound((r->x + r->width*0.5)*scale);
center.y = cvRound((r->y + r->height*0.5)*scale);
radius = cvRound((r->width + r->height)*0.25*scale);
circle( img, center, radius, color, 3, 8, 0 );
} */
//namedWindow("result");
//imshow("result",img);
//waitKey(0);
//destroyAllWindows();
} }
TEST_P(Haar, FaceDetectUseBuf)
{
string imgName = workdir + "lena.jpg";
Mat img = imread( imgName, 1 );
if(img.empty())
{
std::cout << "Couldn't read " << imgName << std::endl;
return ;
}
vector<Rect> faces, oclfaces;
Mat gray, smallImg(cvRound (img.rows / scale), cvRound(img.cols / scale), CV_8UC1 );
MemStorage storage(cvCreateMemStorage(0));
cvtColor( img, gray, CV_BGR2GRAY );
resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
equalizeHist( smallImg, smallImg );
cv::ocl::oclMat image;
image.upload(smallImg);
cascadebuf.detectMultiScale( image, oclfaces, 1.1, 3,
flags,
Size(30, 30), Size(0, 0) );
cascadebuf.release();
cpucascade.detectMultiScale( smallImg, faces, 1.1, 3,
flags,
Size(30, 30), Size(0, 0) );
EXPECT_EQ(faces.size(), oclfaces.size());
}
INSTANTIATE_TEST_CASE_P(FaceDetect, Haar,
Combine(Values(1.0),
Values(CV_HAAR_SCALE_IMAGE, 0)));
#endif // HAVE_OPENCL #endif // HAVE_OPENCL