mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-18 14:13:15 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Fix android build warnings

Browse Source
This commit is contained in:
Andrey Kamaev 2012-09-04 17:44:23 +04:00
parent 8325a28d09
commit 07d92d9e5a
4 changed files with 445 additions and 445 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
modules/contrib/src/bowmsctrainer.cpp
View File

@ -81,46 +81,46 @@ Mat BOWMSCTrainer::cluster() const {
return cluster(mergedDescriptors); return cluster(mergedDescriptors);
} }
Mat BOWMSCTrainer::cluster(const Mat& descriptors) const { Mat BOWMSCTrainer::cluster(const Mat& _descriptors) const {
CV_Assert(!descriptors.empty()); CV_Assert(!_descriptors.empty());
// TODO: sort the descriptors before clustering. // TODO: sort the descriptors before clustering.
Mat icovar = Mat::eye(descriptors.cols,descriptors.cols,descriptors.type()); Mat icovar = Mat::eye(_descriptors.cols,_descriptors.cols,_descriptors.type());
vector<Mat> initialCentres; vector<Mat> initialCentres;
initialCentres.push_back(descriptors.row(0)); initialCentres.push_back(_descriptors.row(0));
for (int i = 1; i < descriptors.rows; i++) { for (int i = 1; i < _descriptors.rows; i++) {
double minDist = DBL_MAX; double minDist = DBL_MAX;
for (size_t j = 0; j < initialCentres.size(); j++) { for (size_t j = 0; j < initialCentres.size(); j++) {
minDist = std::min(minDist, minDist = std::min(minDist,
cv::Mahalanobis(descriptors.row(i),initialCentres[j], cv::Mahalanobis(_descriptors.row(i),initialCentres[j],
icovar)); icovar));
} }
if (minDist > clusterSize) if (minDist > clusterSize)
initialCentres.push_back(descriptors.row(i)); initialCentres.push_back(_descriptors.row(i));
} }
std::vector<std::list<cv::Mat> > clusters; std::vector<std::list<cv::Mat> > clusters;
clusters.resize(initialCentres.size()); clusters.resize(initialCentres.size());
for (int i = 0; i < descriptors.rows; i++) { for (int i = 0; i < _descriptors.rows; i++) {
int index = 0; double dist = 0, minDist = DBL_MAX; int index = 0; double dist = 0, minDist = DBL_MAX;
for (size_t j = 0; j < initialCentres.size(); j++) { for (size_t j = 0; j < initialCentres.size(); j++) {
dist = cv::Mahalanobis(descriptors.row(i),initialCentres[j],icovar); dist = cv::Mahalanobis(_descriptors.row(i),initialCentres[j],icovar);
if (dist < minDist) { if (dist < minDist) {
minDist = dist; minDist = dist;
index = (int)j; index = (int)j;
} }
} }
clusters[index].push_back(descriptors.row(i)); clusters[index].push_back(_descriptors.row(i));
} }
// TODO: throw away small clusters. // TODO: throw away small clusters.
Mat vocabulary; Mat vocabulary;
Mat centre = Mat::zeros(1,descriptors.cols,descriptors.type()); Mat centre = Mat::zeros(1,_descriptors.cols,_descriptors.type());
for (size_t i = 0; i < clusters.size(); i++) { for (size_t i = 0; i < clusters.size(); i++) {
centre.setTo(0); centre.setTo(0);
for (std::list<cv::Mat>::iterator Ci = clusters[i].begin(); Ci != clusters[i].end(); Ci++) { for (std::list<cv::Mat>::iterator Ci = clusters[i].begin(); Ci != clusters[i].end(); Ci++) {

26
modules/contrib/src/openfabmap.cpp
View File

@ -445,16 +445,16 @@ FabMap1::~FabMap1() {
} }
void FabMap1::getLikelihoods(const Mat& queryImgDescriptor, void FabMap1::getLikelihoods(const Mat& queryImgDescriptor,
const vector<Mat>& testImgDescriptors, vector<IMatch>& matches) { const vector<Mat>& testImageDescriptors, vector<IMatch>& matches) {
for (size_t i = 0; i < testImgDescriptors.size(); i++) { for (size_t i = 0; i < testImageDescriptors.size(); i++) {
bool zq, zpq, Lzq; bool zq, zpq, Lzq;
double logP = 0; double logP = 0;
for (int q = 0; q < clTree.cols; q++) { for (int q = 0; q < clTree.cols; q++) {
zq = queryImgDescriptor.at<float>(0,q) > 0; zq = queryImgDescriptor.at<float>(0,q) > 0;
zpq = queryImgDescriptor.at<float>(0,pq(q)) > 0; zpq = queryImgDescriptor.at<float>(0,pq(q)) > 0;
Lzq = testImgDescriptors[i].at<float>(0,q) > 0; Lzq = testImageDescriptors[i].at<float>(0,q) > 0;
logP += log((this->*PzGL)(q, zq, zpq, Lzq)); logP += log((this->*PzGL)(q, zq, zpq, Lzq));
@ -490,16 +490,16 @@ FabMapLUT::~FabMapLUT() {
} }
void FabMapLUT::getLikelihoods(const Mat& queryImgDescriptor, void FabMapLUT::getLikelihoods(const Mat& queryImgDescriptor,
const vector<Mat>& testImgDescriptors, vector<IMatch>& matches) { const vector<Mat>& testImageDescriptors, vector<IMatch>& matches) {
double precFactor = (double)pow(10.0, -precision); double precFactor = (double)pow(10.0, -precision);
for (size_t i = 0; i < testImgDescriptors.size(); i++) { for (size_t i = 0; i < testImageDescriptors.size(); i++) {
unsigned long long int logP = 0; unsigned long long int logP = 0;
for (int q = 0; q < clTree.cols; q++) { for (int q = 0; q < clTree.cols; q++) {
logP += table[q][(queryImgDescriptor.at<float>(0,pq(q)) > 0) + logP += table[q][(queryImgDescriptor.at<float>(0,pq(q)) > 0) +
((queryImgDescriptor.at<float>(0, q) > 0) << 1) + ((queryImgDescriptor.at<float>(0, q) > 0) << 1) +
((testImgDescriptors[i].at<float>(0,q) > 0) << 2)]; ((testImageDescriptors[i].at<float>(0,q) > 0) << 2)];
} }
matches.push_back(IMatch(0,(int)i,-precFactor*(double)logP,0)); matches.push_back(IMatch(0,(int)i,-precFactor*(double)logP,0));
} }
@ -518,7 +518,7 @@ FabMapFBO::~FabMapFBO() {
} }
void FabMapFBO::getLikelihoods(const Mat& queryImgDescriptor, void FabMapFBO::getLikelihoods(const Mat& queryImgDescriptor,
const vector<Mat>& testImgDescriptors, vector<IMatch>& matches) { const vector<Mat>& testImageDescriptors, vector<IMatch>& matches) {
std::multiset<WordStats> wordData; std::multiset<WordStats> wordData;
setWordStatistics(queryImgDescriptor, wordData); setWordStatistics(queryImgDescriptor, wordData);
@ -526,7 +526,7 @@ void FabMapFBO::getLikelihoods(const Mat& queryImgDescriptor,
vector<int> matchIndices; vector<int> matchIndices;
vector<IMatch> queryMatches; vector<IMatch> queryMatches;
for (size_t i = 0; i < testImgDescriptors.size(); i++) { for (size_t i = 0; i < testImageDescriptors.size(); i++) {
queryMatches.push_back(IMatch(0,(int)i,0,0)); queryMatches.push_back(IMatch(0,(int)i,0,0));
matchIndices.push_back((int)i); matchIndices.push_back((int)i);
} }
@ -543,7 +543,7 @@ void FabMapFBO::getLikelihoods(const Mat& queryImgDescriptor,
for (size_t i = 0; i < matchIndices.size(); i++) { for (size_t i = 0; i < matchIndices.size(); i++) {
bool Lzq = bool Lzq =
testImgDescriptors[matchIndices[i]].at<float>(0,wordIter->q) > 0; testImageDescriptors[matchIndices[i]].at<float>(0,wordIter->q) > 0;
queryMatches[matchIndices[i]].likelihood += queryMatches[matchIndices[i]].likelihood +=
log((this->*PzGL)(wordIter->q,zq,zpq,Lzq)); log((this->*PzGL)(wordIter->q,zq,zpq,Lzq));
currBest = currBest =
@ -689,17 +689,17 @@ void FabMap2::add(const vector<Mat>& queryImgDescriptors) {
} }
void FabMap2::getLikelihoods(const Mat& queryImgDescriptor, void FabMap2::getLikelihoods(const Mat& queryImgDescriptor,
const vector<Mat>& testImgDescriptors, vector<IMatch>& matches) { const vector<Mat>& testImageDescriptors, vector<IMatch>& matches) {
if (&testImgDescriptors== &(this->testImgDescriptors)) { if (&testImageDescriptors == &testImgDescriptors) {
getIndexLikelihoods(queryImgDescriptor, testDefaults, testInvertedMap, getIndexLikelihoods(queryImgDescriptor, testDefaults, testInvertedMap,
matches); matches);
} else { } else {
CV_Assert(!(flags & MOTION_MODEL)); CV_Assert(!(flags & MOTION_MODEL));
vector<double> defaults; vector<double> defaults;
std::map<int, vector<int> > invertedMap; std::map<int, vector<int> > invertedMap;
for (size_t i = 0; i < testImgDescriptors.size(); i++) { for (size_t i = 0; i < testImageDescriptors.size(); i++) {
addToIndex(testImgDescriptors[i],defaults,invertedMap); addToIndex(testImageDescriptors[i],defaults,invertedMap);
} }
getIndexLikelihoods(queryImgDescriptor, defaults, invertedMap, matches); getIndexLikelihoods(queryImgDescriptor, defaults, invertedMap, matches);
} }

742
modules/objdetect/src/haar.cpp
View File

@ -47,18 +47,18 @@
#if CV_SSE2 || CV_SSE3 #if CV_SSE2 || CV_SSE3
# if !CV_SSE4_1 && !CV_SSE4_2 # if !CV_SSE4_1 && !CV_SSE4_2
# define _mm_blendv_pd(a, b, m) _mm_xor_pd(a, _mm_and_pd(_mm_xor_pd(b, a), m)) # define _mm_blendv_pd(a, b, m) _mm_xor_pd(a, _mm_and_pd(_mm_xor_pd(b, a), m))
# define _mm_blendv_ps(a, b, m) _mm_xor_ps(a, _mm_and_ps(_mm_xor_ps(b, a), m)) # define _mm_blendv_ps(a, b, m) _mm_xor_ps(a, _mm_and_ps(_mm_xor_ps(b, a), m))
# endif # endif
#endif #endif
# if CV_AVX # if CV_AVX
# define CV_HAAR_USE_AVX 1 # define CV_HAAR_USE_AVX 1
# else # else
# if CV_SSE2 || CV_SSE3 # if CV_SSE2 || CV_SSE3
# define CV_HAAR_USE_SSE 1 # define CV_HAAR_USE_SSE 1
# endif # endif
# endif # endif
/* these settings affect the quality of detection: change with care */ /* these settings affect the quality of detection: change with care */
#define CV_ADJUST_FEATURES 1 #define CV_ADJUST_FEATURES 1
@ -637,83 +637,83 @@ cvSetImagesForHaarClassifierCascade( CvHaarClassifierCascade* _cascade,
#ifdef CV_HAAR_USE_AVX #ifdef CV_HAAR_USE_AVX
CV_INLINE CV_INLINE
double icvEvalHidHaarClassifierAVX( CvHidHaarClassifier* classifier, double icvEvalHidHaarClassifierAVX( CvHidHaarClassifier* classifier,
double variance_norm_factor, size_t p_offset ) double variance_norm_factor, size_t p_offset )
{ {
int CV_DECL_ALIGNED(32) idxV[8] = {0,0,0,0,0,0,0,0}; int CV_DECL_ALIGNED(32) idxV[8] = {0,0,0,0,0,0,0,0};
char flags[8] = {0,0,0,0,0,0,0,0}; char flags[8] = {0,0,0,0,0,0,0,0};
CvHidHaarTreeNode* nodes[8]; CvHidHaarTreeNode* nodes[8];
double res = 0; double res = 0;
char exitConditionFlag = 0; char exitConditionFlag = 0;
for(;;) for(;;)
{ {
float CV_DECL_ALIGNED(32) tmp[8] = {0,0,0,0,0,0,0,0}; float CV_DECL_ALIGNED(32) tmp[8] = {0,0,0,0,0,0,0,0};
nodes[0] = classifier ->node + idxV[0]; nodes[0] = classifier ->node + idxV[0];
nodes[1] = (classifier+1)->node + idxV[1]; nodes[1] = (classifier+1)->node + idxV[1];
nodes[2] = (classifier+2)->node + idxV[2]; nodes[2] = (classifier+2)->node + idxV[2];
nodes[3] = (classifier+3)->node + idxV[3]; nodes[3] = (classifier+3)->node + idxV[3];
nodes[4] = (classifier+4)->node + idxV[4]; nodes[4] = (classifier+4)->node + idxV[4];
nodes[5] = (classifier+5)->node + idxV[5]; nodes[5] = (classifier+5)->node + idxV[5];
nodes[6] = (classifier+6)->node + idxV[6]; nodes[6] = (classifier+6)->node + idxV[6];
nodes[7] = (classifier+7)->node + idxV[7]; nodes[7] = (classifier+7)->node + idxV[7];
__m256 t = _mm256_set1_ps(variance_norm_factor); __m256 t = _mm256_set1_ps(variance_norm_factor);
t = _mm256_mul_ps(t, _mm256_set_ps(nodes[7]->threshold,nodes[6]->threshold,nodes[5]->threshold,nodes[4]->threshold,nodes[3]->threshold,nodes[2]->threshold,nodes[1]->threshold,nodes[0]->threshold)); t = _mm256_mul_ps(t, _mm256_set_ps(nodes[7]->threshold,nodes[6]->threshold,nodes[5]->threshold,nodes[4]->threshold,nodes[3]->threshold,nodes[2]->threshold,nodes[1]->threshold,nodes[0]->threshold));
__m256 offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[0],p_offset), calc_sum(nodes[6]->feature.rect[0],p_offset), calc_sum(nodes[5]->feature.rect[0],p_offset), __m256 offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[0],p_offset), calc_sum(nodes[6]->feature.rect[0],p_offset), calc_sum(nodes[5]->feature.rect[0],p_offset),
calc_sum(nodes[4]->feature.rect[0],p_offset), calc_sum(nodes[3]->feature.rect[0],p_offset), calc_sum(nodes[2]->feature.rect[0],p_offset), calc_sum(nodes[1]->feature.rect[0], calc_sum(nodes[4]->feature.rect[0],p_offset), calc_sum(nodes[3]->feature.rect[0],p_offset), calc_sum(nodes[2]->feature.rect[0],p_offset), calc_sum(nodes[1]->feature.rect[0],
p_offset),calc_sum(nodes[0]->feature.rect[0],p_offset)); p_offset),calc_sum(nodes[0]->feature.rect[0],p_offset));
__m256 weight = _mm256_set_ps(nodes[7]->feature.rect[0].weight, nodes[6]->feature.rect[0].weight, nodes[5]->feature.rect[0].weight, __m256 weight = _mm256_set_ps(nodes[7]->feature.rect[0].weight, nodes[6]->feature.rect[0].weight, nodes[5]->feature.rect[0].weight,
nodes[4]->feature.rect[0].weight, nodes[3]->feature.rect[0].weight, nodes[2]->feature.rect[0].weight, nodes[1]->feature.rect[0].weight, nodes[0]->feature.rect[0].weight); nodes[4]->feature.rect[0].weight, nodes[3]->feature.rect[0].weight, nodes[2]->feature.rect[0].weight, nodes[1]->feature.rect[0].weight, nodes[0]->feature.rect[0].weight);
__m256 sum = _mm256_mul_ps(offset, weight); __m256 sum = _mm256_mul_ps(offset, weight);
offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[1],p_offset),calc_sum(nodes[6]->feature.rect[1],p_offset),calc_sum(nodes[5]->feature.rect[1],p_offset), offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[1],p_offset),calc_sum(nodes[6]->feature.rect[1],p_offset),calc_sum(nodes[5]->feature.rect[1],p_offset),
calc_sum(nodes[4]->feature.rect[1],p_offset),calc_sum(nodes[3]->feature.rect[1],p_offset),calc_sum(nodes[2]->feature.rect[1],p_offset),calc_sum(nodes[1]->feature.rect[1],p_offset), calc_sum(nodes[4]->feature.rect[1],p_offset),calc_sum(nodes[3]->feature.rect[1],p_offset),calc_sum(nodes[2]->feature.rect[1],p_offset),calc_sum(nodes[1]->feature.rect[1],p_offset),
calc_sum(nodes[0]->feature.rect[1],p_offset)); calc_sum(nodes[0]->feature.rect[1],p_offset));
weight = _mm256_set_ps(nodes[7]->feature.rect[1].weight, nodes[6]->feature.rect[1].weight, nodes[5]->feature.rect[1].weight, nodes[4]->feature.rect[1].weight, weight = _mm256_set_ps(nodes[7]->feature.rect[1].weight, nodes[6]->feature.rect[1].weight, nodes[5]->feature.rect[1].weight, nodes[4]->feature.rect[1].weight,
nodes[3]->feature.rect[1].weight, nodes[2]->feature.rect[1].weight, nodes[1]->feature.rect[1].weight, nodes[0]->feature.rect[1].weight); nodes[3]->feature.rect[1].weight, nodes[2]->feature.rect[1].weight, nodes[1]->feature.rect[1].weight, nodes[0]->feature.rect[1].weight);
sum = _mm256_add_ps(sum, _mm256_mul_ps(offset,weight)); sum = _mm256_add_ps(sum, _mm256_mul_ps(offset,weight));
if( nodes[0]->feature.rect[2].p0 ) if( nodes[0]->feature.rect[2].p0 )
tmp[0] = calc_sum(nodes[0]->feature.rect[2],p_offset) * nodes[0]->feature.rect[2].weight; tmp[0] = calc_sum(nodes[0]->feature.rect[2],p_offset) * nodes[0]->feature.rect[2].weight;
if( nodes[1]->feature.rect[2].p0 ) if( nodes[1]->feature.rect[2].p0 )
tmp[1] = calc_sum(nodes[1]->feature.rect[2],p_offset) * nodes[1]->feature.rect[2].weight; tmp[1] = calc_sum(nodes[1]->feature.rect[2],p_offset) * nodes[1]->feature.rect[2].weight;
if( nodes[2]->feature.rect[2].p0 ) if( nodes[2]->feature.rect[2].p0 )
tmp[2] = calc_sum(nodes[2]->feature.rect[2],p_offset) * nodes[2]->feature.rect[2].weight; tmp[2] = calc_sum(nodes[2]->feature.rect[2],p_offset) * nodes[2]->feature.rect[2].weight;
if( nodes[3]->feature.rect[2].p0 ) if( nodes[3]->feature.rect[2].p0 )
tmp[3] = calc_sum(nodes[3]->feature.rect[2],p_offset) * nodes[3]->feature.rect[2].weight; tmp[3] = calc_sum(nodes[3]->feature.rect[2],p_offset) * nodes[3]->feature.rect[2].weight;
if( nodes[4]->feature.rect[2].p0 ) if( nodes[4]->feature.rect[2].p0 )
tmp[4] = calc_sum(nodes[4]->feature.rect[2],p_offset) * nodes[4]->feature.rect[2].weight; tmp[4] = calc_sum(nodes[4]->feature.rect[2],p_offset) * nodes[4]->feature.rect[2].weight;
if( nodes[5]->feature.rect[2].p0 ) if( nodes[5]->feature.rect[2].p0 )
tmp[5] = calc_sum(nodes[5]->feature.rect[2],p_offset) * nodes[5]->feature.rect[2].weight; tmp[5] = calc_sum(nodes[5]->feature.rect[2],p_offset) * nodes[5]->feature.rect[2].weight;
if( nodes[6]->feature.rect[2].p0 ) if( nodes[6]->feature.rect[2].p0 )
tmp[6] = calc_sum(nodes[6]->feature.rect[2],p_offset) * nodes[6]->feature.rect[2].weight; tmp[6] = calc_sum(nodes[6]->feature.rect[2],p_offset) * nodes[6]->feature.rect[2].weight;
if( nodes[7]->feature.rect[2].p0 ) if( nodes[7]->feature.rect[2].p0 )
tmp[7] = calc_sum(nodes[7]->feature.rect[2],p_offset) * nodes[7]->feature.rect[2].weight; tmp[7] = calc_sum(nodes[7]->feature.rect[2],p_offset) * nodes[7]->feature.rect[2].weight;
sum = _mm256_add_ps(sum,_mm256_load_ps(tmp)); sum = _mm256_add_ps(sum,_mm256_load_ps(tmp));
__m256 left = _mm256_set_ps(nodes[7]->left,nodes[6]->left,nodes[5]->left,nodes[4]->left,nodes[3]->left,nodes[2]->left,nodes[1]->left,nodes[0]->left); __m256 left = _mm256_set_ps(nodes[7]->left,nodes[6]->left,nodes[5]->left,nodes[4]->left,nodes[3]->left,nodes[2]->left,nodes[1]->left,nodes[0]->left);
__m256 right = _mm256_set_ps(nodes[7]->right,nodes[6]->right,nodes[5]->right,nodes[4]->right,nodes[3]->right,nodes[2]->right,nodes[1]->right,nodes[0]->right); __m256 right = _mm256_set_ps(nodes[7]->right,nodes[6]->right,nodes[5]->right,nodes[4]->right,nodes[3]->right,nodes[2]->right,nodes[1]->right,nodes[0]->right);
_mm256_store_si256((__m256i*)idxV,_mm256_cvttps_epi32(_mm256_blendv_ps(right, left,_mm256_cmp_ps(sum, t, _CMP_LT_OQ )))); _mm256_store_si256((__m256i*)idxV,_mm256_cvttps_epi32(_mm256_blendv_ps(right, left,_mm256_cmp_ps(sum, t, _CMP_LT_OQ ))));
for(int i = 0; i < 8; i++) for(int i = 0; i < 8; i++)
{ {
if(idxV[i]<=0) if(idxV[i]<=0)
{ {
if(!flags[i]) if(!flags[i])
{ {
exitConditionFlag++; exitConditionFlag++;
flags[i]=1; flags[i]=1;
res+=((classifier+i)->alpha[-idxV[i]]); res+=((classifier+i)->alpha[-idxV[i]]);
} }
idxV[i]=0; idxV[i]=0;
} }
} }
if(exitConditionFlag==8) if(exitConditionFlag==8)
return res; return res;
} }
} }
#endif #endif
@ -723,50 +723,50 @@ double icvEvalHidHaarClassifier( CvHidHaarClassifier* classifier,
size_t p_offset ) size_t p_offset )
{ {
int idx = 0; int idx = 0;
/*#if CV_HAAR_USE_SSE && !CV_HAAR_USE_AVX /*#if CV_HAAR_USE_SSE && !CV_HAAR_USE_AVX
if(cv::checkHardwareSupport(CV_CPU_SSE2))//based on old SSE variant. Works slow if(cv::checkHardwareSupport(CV_CPU_SSE2))//based on old SSE variant. Works slow
{ {
double CV_DECL_ALIGNED(16) temp[2]; double CV_DECL_ALIGNED(16) temp[2];
__m128d zero = _mm_setzero_pd(); __m128d zero = _mm_setzero_pd();
do do
{ {
CvHidHaarTreeNode* node = classifier->node + idx; CvHidHaarTreeNode* node = classifier->node + idx;
__m128d t = _mm_set1_pd((node->threshold)*variance_norm_factor); __m128d t = _mm_set1_pd((node->threshold)*variance_norm_factor);
__m128d left = _mm_set1_pd(node->left); __m128d left = _mm_set1_pd(node->left);
__m128d right = _mm_set1_pd(node->right); __m128d right = _mm_set1_pd(node->right);
double _sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double _sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
_sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; _sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
if( node->feature.rect[2].p0 ) if( node->feature.rect[2].p0 )
_sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight; _sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
__m128d sum = _mm_set1_pd(_sum); __m128d sum = _mm_set1_pd(_sum);
t = _mm_cmplt_sd(sum, t); t = _mm_cmplt_sd(sum, t);
sum = _mm_blendv_pd(right, left, t); sum = _mm_blendv_pd(right, left, t);
_mm_store_pd(temp, sum); _mm_store_pd(temp, sum);
idx = (int)temp[0]; idx = (int)temp[0];
} }
while(idx > 0 ); while(idx > 0 );
} }
else else
#endif*/ #endif*/
{ {
do do
{ {
CvHidHaarTreeNode* node = classifier->node + idx; CvHidHaarTreeNode* node = classifier->node + idx;
double t = node->threshold * variance_norm_factor; double t = node->threshold * variance_norm_factor;
double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
if( node->feature.rect[2].p0 ) if( node->feature.rect[2].p0 )
sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight; sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
idx = sum < t ? node->left : node->right; idx = sum < t ? node->left : node->right;
} }
while( idx > 0 ); while( idx > 0 );
} }
return classifier->alpha[-idx]; return classifier->alpha[-idx];
} }
@ -777,18 +777,18 @@ static int
cvRunHaarClassifierCascadeSum( const CvHaarClassifierCascade* _cascade, cvRunHaarClassifierCascadeSum( const CvHaarClassifierCascade* _cascade,
CvPoint pt, double& stage_sum, int start_stage ) CvPoint pt, double& stage_sum, int start_stage )
{ {
#ifdef CV_HAAR_USE_AVX #ifdef CV_HAAR_USE_AVX
bool haveAVX = false; bool haveAVX = false;
if(cv::checkHardwareSupport(CV_CPU_AVX)) if(cv::checkHardwareSupport(CV_CPU_AVX))
if(_xgetbv(_XCR_XFEATURE_ENABLED_MASK)&0x6)// Check if the OS will save the YMM registers if(_xgetbv(_XCR_XFEATURE_ENABLED_MASK)&0x6)// Check if the OS will save the YMM registers
{ {
haveAVX = true; haveAVX = true;
} }
#else #else
#ifdef CV_HAAR_USE_SSE #ifdef CV_HAAR_USE_SSE
bool haveSSE2 = cv::checkHardwareSupport(CV_CPU_SSE2); bool haveSSE2 = cv::checkHardwareSupport(CV_CPU_SSE2);
#endif #endif
#endif #endif
int p_offset, pq_offset; int p_offset, pq_offset;
int i, j; int i, j;
@ -828,17 +828,17 @@ cvRunHaarClassifierCascadeSum( const CvHaarClassifierCascade* _cascade,
{ {
stage_sum = 0.0; stage_sum = 0.0;
#ifdef CV_HAAR_USE_AVX #ifdef CV_HAAR_USE_AVX
if(haveAVX) if(haveAVX)
{ {
for( ; j < cascade->stage_classifier[i].count-8; j+=8 ) for( ; j < cascade->stage_classifier[i].count-8; j+=8 )
{ {
stage_sum += icvEvalHidHaarClassifierAVX( stage_sum += icvEvalHidHaarClassifierAVX(
cascade->stage_classifier[i].classifier+j, cascade->stage_classifier[i].classifier+j,
variance_norm_factor, p_offset ); variance_norm_factor, p_offset );
} }
} }
#endif #endif
for( j = 0; j < ptr->count; j++ ) for( j = 0; j < ptr->count; j++ )
{ {
stage_sum += icvEvalHidHaarClassifier( ptr->classifier + j, variance_norm_factor, p_offset ); stage_sum += icvEvalHidHaarClassifier( ptr->classifier + j, variance_norm_factor, p_offset );
@ -859,283 +859,283 @@ cvRunHaarClassifierCascadeSum( const CvHaarClassifierCascade* _cascade,
} }
else if( cascade->isStumpBased ) else if( cascade->isStumpBased )
{ {
#ifdef CV_HAAR_USE_AVX #ifdef CV_HAAR_USE_AVX
if(haveAVX) if(haveAVX)
{ {
CvHidHaarClassifier* classifiers[8]; CvHidHaarClassifier* classifiers[8];
CvHidHaarTreeNode* nodes[8]; CvHidHaarTreeNode* nodes[8];
for( i = start_stage; i < cascade->count; i++ ) for( i = start_stage; i < cascade->count; i++ )
{ {
stage_sum = 0.0; stage_sum = 0.0;
int j = 0; int j = 0;
float CV_DECL_ALIGNED(32) buf[8]; float CV_DECL_ALIGNED(32) buf[8];
if( cascade->stage_classifier[i].two_rects ) if( cascade->stage_classifier[i].two_rects )
{ {
for( ; j <= cascade->stage_classifier[i].count-8; j+=8 ) for( ; j <= cascade->stage_classifier[i].count-8; j+=8 )
{ {
//__m256 stage_sumPart = _mm256_setzero_ps(); //__m256 stage_sumPart = _mm256_setzero_ps();
classifiers[0] = cascade->stage_classifier[i].classifier + j; classifiers[0] = cascade->stage_classifier[i].classifier + j;
nodes[0] = classifiers[0]->node; nodes[0] = classifiers[0]->node;
classifiers[1] = cascade->stage_classifier[i].classifier + j + 1; classifiers[1] = cascade->stage_classifier[i].classifier + j + 1;
nodes[1] = classifiers[1]->node; nodes[1] = classifiers[1]->node;
classifiers[2] = cascade->stage_classifier[i].classifier + j + 2; classifiers[2] = cascade->stage_classifier[i].classifier + j + 2;
nodes[2]= classifiers[2]->node; nodes[2]= classifiers[2]->node;
classifiers[3] = cascade->stage_classifier[i].classifier + j + 3; classifiers[3] = cascade->stage_classifier[i].classifier + j + 3;
nodes[3] = classifiers[3]->node; nodes[3] = classifiers[3]->node;
classifiers[4] = cascade->stage_classifier[i].classifier + j + 4; classifiers[4] = cascade->stage_classifier[i].classifier + j + 4;
nodes[4] = classifiers[4]->node; nodes[4] = classifiers[4]->node;
classifiers[5] = cascade->stage_classifier[i].classifier + j + 5; classifiers[5] = cascade->stage_classifier[i].classifier + j + 5;
nodes[5] = classifiers[5]->node; nodes[5] = classifiers[5]->node;
classifiers[6] = cascade->stage_classifier[i].classifier + j + 6; classifiers[6] = cascade->stage_classifier[i].classifier + j + 6;
nodes[6] = classifiers[6]->node; nodes[6] = classifiers[6]->node;
classifiers[7] = cascade->stage_classifier[i].classifier + j + 7; classifiers[7] = cascade->stage_classifier[i].classifier + j + 7;
nodes[7] = classifiers[7]->node; nodes[7] = classifiers[7]->node;
__m256 t = _mm256_set1_ps(variance_norm_factor); __m256 t = _mm256_set1_ps(variance_norm_factor);
t = _mm256_mul_ps(t, _mm256_set_ps(nodes[7]->threshold,nodes[6]->threshold,nodes[5]->threshold,nodes[4]->threshold,nodes[3]->threshold,nodes[2]->threshold,nodes[1]->threshold,nodes[0]->threshold)); t = _mm256_mul_ps(t, _mm256_set_ps(nodes[7]->threshold,nodes[6]->threshold,nodes[5]->threshold,nodes[4]->threshold,nodes[3]->threshold,nodes[2]->threshold,nodes[1]->threshold,nodes[0]->threshold));
__m256 offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[0],p_offset), calc_sum(nodes[6]->feature.rect[0],p_offset), calc_sum(nodes[5]->feature.rect[0],p_offset), __m256 offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[0],p_offset), calc_sum(nodes[6]->feature.rect[0],p_offset), calc_sum(nodes[5]->feature.rect[0],p_offset),
calc_sum(nodes[4]->feature.rect[0],p_offset), calc_sum(nodes[3]->feature.rect[0],p_offset), calc_sum(nodes[2]->feature.rect[0],p_offset), calc_sum(nodes[1]->feature.rect[0], calc_sum(nodes[4]->feature.rect[0],p_offset), calc_sum(nodes[3]->feature.rect[0],p_offset), calc_sum(nodes[2]->feature.rect[0],p_offset), calc_sum(nodes[1]->feature.rect[0],
p_offset),calc_sum(nodes[0]->feature.rect[0],p_offset)); p_offset),calc_sum(nodes[0]->feature.rect[0],p_offset));
__m256 weight = _mm256_set_ps(nodes[7]->feature.rect[0].weight, nodes[6]->feature.rect[0].weight, nodes[5]->feature.rect[0].weight, __m256 weight = _mm256_set_ps(nodes[7]->feature.rect[0].weight, nodes[6]->feature.rect[0].weight, nodes[5]->feature.rect[0].weight,
nodes[4]->feature.rect[0].weight, nodes[3]->feature.rect[0].weight, nodes[2]->feature.rect[0].weight, nodes[1]->feature.rect[0].weight, nodes[0]->feature.rect[0].weight); nodes[4]->feature.rect[0].weight, nodes[3]->feature.rect[0].weight, nodes[2]->feature.rect[0].weight, nodes[1]->feature.rect[0].weight, nodes[0]->feature.rect[0].weight);
__m256 sum = _mm256_mul_ps(offset, weight); __m256 sum = _mm256_mul_ps(offset, weight);
offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[1],p_offset),calc_sum(nodes[6]->feature.rect[1],p_offset),calc_sum(nodes[5]->feature.rect[1],p_offset), offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[1],p_offset),calc_sum(nodes[6]->feature.rect[1],p_offset),calc_sum(nodes[5]->feature.rect[1],p_offset),
calc_sum(nodes[4]->feature.rect[1],p_offset),calc_sum(nodes[3]->feature.rect[1],p_offset),calc_sum(nodes[2]->feature.rect[1],p_offset),calc_sum(nodes[1]->feature.rect[1],p_offset), calc_sum(nodes[4]->feature.rect[1],p_offset),calc_sum(nodes[3]->feature.rect[1],p_offset),calc_sum(nodes[2]->feature.rect[1],p_offset),calc_sum(nodes[1]->feature.rect[1],p_offset),
calc_sum(nodes[0]->feature.rect[1],p_offset)); calc_sum(nodes[0]->feature.rect[1],p_offset));
weight = _mm256_set_ps(nodes[7]->feature.rect[1].weight, nodes[6]->feature.rect[1].weight, nodes[5]->feature.rect[1].weight, nodes[4]->feature.rect[1].weight, weight = _mm256_set_ps(nodes[7]->feature.rect[1].weight, nodes[6]->feature.rect[1].weight, nodes[5]->feature.rect[1].weight, nodes[4]->feature.rect[1].weight,
nodes[3]->feature.rect[1].weight, nodes[2]->feature.rect[1].weight, nodes[1]->feature.rect[1].weight, nodes[0]->feature.rect[1].weight); nodes[3]->feature.rect[1].weight, nodes[2]->feature.rect[1].weight, nodes[1]->feature.rect[1].weight, nodes[0]->feature.rect[1].weight);
sum = _mm256_add_ps(sum, _mm256_mul_ps(offset,weight)); sum = _mm256_add_ps(sum, _mm256_mul_ps(offset,weight));
__m256 alpha0 = _mm256_set_ps(classifiers[7]->alpha[0],classifiers[6]->alpha[0],classifiers[5]->alpha[0],classifiers[4]->alpha[0],classifiers[3]->alpha[0], __m256 alpha0 = _mm256_set_ps(classifiers[7]->alpha[0],classifiers[6]->alpha[0],classifiers[5]->alpha[0],classifiers[4]->alpha[0],classifiers[3]->alpha[0],
classifiers[2]->alpha[0],classifiers[1]->alpha[0],classifiers[0]->alpha[0]); classifiers[2]->alpha[0],classifiers[1]->alpha[0],classifiers[0]->alpha[0]);
__m256 alpha1 = _mm256_set_ps(classifiers[7]->alpha[1],classifiers[6]->alpha[1],classifiers[5]->alpha[1],classifiers[4]->alpha[1],classifiers[3]->alpha[1], __m256 alpha1 = _mm256_set_ps(classifiers[7]->alpha[1],classifiers[6]->alpha[1],classifiers[5]->alpha[1],classifiers[4]->alpha[1],classifiers[3]->alpha[1],
classifiers[2]->alpha[1],classifiers[1]->alpha[1],classifiers[0]->alpha[1]); classifiers[2]->alpha[1],classifiers[1]->alpha[1],classifiers[0]->alpha[1]);
_mm256_store_ps(buf, _mm256_blendv_ps(alpha0, alpha1, _mm256_cmp_ps(t, sum, _CMP_LE_OQ ))); _mm256_store_ps(buf, _mm256_blendv_ps(alpha0, alpha1, _mm256_cmp_ps(t, sum, _CMP_LE_OQ )));
stage_sum+=(buf[0]+buf[1]+buf[2]+buf[3]+buf[4]+buf[5]+buf[6]+buf[7]); stage_sum+=(buf[0]+buf[1]+buf[2]+buf[3]+buf[4]+buf[5]+buf[6]+buf[7]);
} }
for( ; j < cascade->stage_classifier[i].count; j++ ) for( ; j < cascade->stage_classifier[i].count; j++ )
{ {
CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j; CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
CvHidHaarTreeNode* node = classifier->node; CvHidHaarTreeNode* node = classifier->node;
double t = node->threshold*variance_norm_factor; double t = node->threshold*variance_norm_factor;
double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
stage_sum += classifier->alpha[sum >= t]; stage_sum += classifier->alpha[sum >= t];
} }
} }
else else
{ {
for( ; j <= (cascade->stage_classifier[i].count)-8; j+=8 ) for( ; j <= (cascade->stage_classifier[i].count)-8; j+=8 )
{ {
float CV_DECL_ALIGNED(32) tmp[8] = {0,0,0,0,0,0,0,0}; float CV_DECL_ALIGNED(32) tmp[8] = {0,0,0,0,0,0,0,0};
classifiers[0] = cascade->stage_classifier[i].classifier + j; classifiers[0] = cascade->stage_classifier[i].classifier + j;
nodes[0] = classifiers[0]->node; nodes[0] = classifiers[0]->node;
classifiers[1] = cascade->stage_classifier[i].classifier + j + 1; classifiers[1] = cascade->stage_classifier[i].classifier + j + 1;
nodes[1] = classifiers[1]->node; nodes[1] = classifiers[1]->node;
classifiers[2] = cascade->stage_classifier[i].classifier + j + 2; classifiers[2] = cascade->stage_classifier[i].classifier + j + 2;
nodes[2]= classifiers[2]->node; nodes[2]= classifiers[2]->node;
classifiers[3] = cascade->stage_classifier[i].classifier + j + 3; classifiers[3] = cascade->stage_classifier[i].classifier + j + 3;
nodes[3] = classifiers[3]->node; nodes[3] = classifiers[3]->node;
classifiers[4] = cascade->stage_classifier[i].classifier + j + 4; classifiers[4] = cascade->stage_classifier[i].classifier + j + 4;
nodes[4] = classifiers[4]->node; nodes[4] = classifiers[4]->node;
classifiers[5] = cascade->stage_classifier[i].classifier + j + 5; classifiers[5] = cascade->stage_classifier[i].classifier + j + 5;
nodes[5] = classifiers[5]->node; nodes[5] = classifiers[5]->node;
classifiers[6] = cascade->stage_classifier[i].classifier + j + 6; classifiers[6] = cascade->stage_classifier[i].classifier + j + 6;
nodes[6] = classifiers[6]->node; nodes[6] = classifiers[6]->node;
classifiers[7] = cascade->stage_classifier[i].classifier + j + 7; classifiers[7] = cascade->stage_classifier[i].classifier + j + 7;
nodes[7] = classifiers[7]->node; nodes[7] = classifiers[7]->node;
__m256 t = _mm256_set1_ps(variance_norm_factor); __m256 t = _mm256_set1_ps(variance_norm_factor);
t = _mm256_mul_ps(t, _mm256_set_ps(nodes[7]->threshold,nodes[6]->threshold,nodes[5]->threshold,nodes[4]->threshold,nodes[3]->threshold,nodes[2]->threshold,nodes[1]->threshold,nodes[0]->threshold)); t = _mm256_mul_ps(t, _mm256_set_ps(nodes[7]->threshold,nodes[6]->threshold,nodes[5]->threshold,nodes[4]->threshold,nodes[3]->threshold,nodes[2]->threshold,nodes[1]->threshold,nodes[0]->threshold));
__m256 offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[0],p_offset), calc_sum(nodes[6]->feature.rect[0],p_offset), calc_sum(nodes[5]->feature.rect[0],p_offset), __m256 offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[0],p_offset), calc_sum(nodes[6]->feature.rect[0],p_offset), calc_sum(nodes[5]->feature.rect[0],p_offset),
calc_sum(nodes[4]->feature.rect[0],p_offset), calc_sum(nodes[3]->feature.rect[0],p_offset), calc_sum(nodes[2]->feature.rect[0],p_offset), calc_sum(nodes[1]->feature.rect[0], calc_sum(nodes[4]->feature.rect[0],p_offset), calc_sum(nodes[3]->feature.rect[0],p_offset), calc_sum(nodes[2]->feature.rect[0],p_offset), calc_sum(nodes[1]->feature.rect[0],
p_offset),calc_sum(nodes[0]->feature.rect[0],p_offset)); p_offset),calc_sum(nodes[0]->feature.rect[0],p_offset));
__m256 weight = _mm256_set_ps(nodes[7]->feature.rect[0].weight, nodes[6]->feature.rect[0].weight, nodes[5]->feature.rect[0].weight, __m256 weight = _mm256_set_ps(nodes[7]->feature.rect[0].weight, nodes[6]->feature.rect[0].weight, nodes[5]->feature.rect[0].weight,
nodes[4]->feature.rect[0].weight, nodes[3]->feature.rect[0].weight, nodes[2]->feature.rect[0].weight, nodes[1]->feature.rect[0].weight, nodes[0]->feature.rect[0].weight); nodes[4]->feature.rect[0].weight, nodes[3]->feature.rect[0].weight, nodes[2]->feature.rect[0].weight, nodes[1]->feature.rect[0].weight, nodes[0]->feature.rect[0].weight);
__m256 sum = _mm256_mul_ps(offset, weight); __m256 sum = _mm256_mul_ps(offset, weight);
offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[1],p_offset),calc_sum(nodes[6]->feature.rect[1],p_offset),calc_sum(nodes[5]->feature.rect[1],p_offset), offset = _mm256_set_ps(calc_sum(nodes[7]->feature.rect[1],p_offset),calc_sum(nodes[6]->feature.rect[1],p_offset),calc_sum(nodes[5]->feature.rect[1],p_offset),
calc_sum(nodes[4]->feature.rect[1],p_offset),calc_sum(nodes[3]->feature.rect[1],p_offset),calc_sum(nodes[2]->feature.rect[1],p_offset),calc_sum(nodes[1]->feature.rect[1],p_offset), calc_sum(nodes[4]->feature.rect[1],p_offset),calc_sum(nodes[3]->feature.rect[1],p_offset),calc_sum(nodes[2]->feature.rect[1],p_offset),calc_sum(nodes[1]->feature.rect[1],p_offset),
calc_sum(nodes[0]->feature.rect[1],p_offset)); calc_sum(nodes[0]->feature.rect[1],p_offset));
weight = _mm256_set_ps(nodes[7]->feature.rect[1].weight, nodes[6]->feature.rect[1].weight, nodes[5]->feature.rect[1].weight, nodes[4]->feature.rect[1].weight, weight = _mm256_set_ps(nodes[7]->feature.rect[1].weight, nodes[6]->feature.rect[1].weight, nodes[5]->feature.rect[1].weight, nodes[4]->feature.rect[1].weight,
nodes[3]->feature.rect[1].weight, nodes[2]->feature.rect[1].weight, nodes[1]->feature.rect[1].weight, nodes[0]->feature.rect[1].weight); nodes[3]->feature.rect[1].weight, nodes[2]->feature.rect[1].weight, nodes[1]->feature.rect[1].weight, nodes[0]->feature.rect[1].weight);
sum = _mm256_add_ps(sum, _mm256_mul_ps(offset,weight)); sum = _mm256_add_ps(sum, _mm256_mul_ps(offset,weight));
if( nodes[0]->feature.rect[2].p0 ) if( nodes[0]->feature.rect[2].p0 )
tmp[0] = calc_sum(nodes[0]->feature.rect[2],p_offset) * nodes[0]->feature.rect[2].weight; tmp[0] = calc_sum(nodes[0]->feature.rect[2],p_offset) * nodes[0]->feature.rect[2].weight;
if( nodes[1]->feature.rect[2].p0 ) if( nodes[1]->feature.rect[2].p0 )
tmp[1] = calc_sum(nodes[1]->feature.rect[2],p_offset) * nodes[1]->feature.rect[2].weight; tmp[1] = calc_sum(nodes[1]->feature.rect[2],p_offset) * nodes[1]->feature.rect[2].weight;
if( nodes[2]->feature.rect[2].p0 ) if( nodes[2]->feature.rect[2].p0 )
tmp[2] = calc_sum(nodes[2]->feature.rect[2],p_offset) * nodes[2]->feature.rect[2].weight; tmp[2] = calc_sum(nodes[2]->feature.rect[2],p_offset) * nodes[2]->feature.rect[2].weight;
if( nodes[3]->feature.rect[2].p0 ) if( nodes[3]->feature.rect[2].p0 )
tmp[3] = calc_sum(nodes[3]->feature.rect[2],p_offset) * nodes[3]->feature.rect[2].weight; tmp[3] = calc_sum(nodes[3]->feature.rect[2],p_offset) * nodes[3]->feature.rect[2].weight;
if( nodes[4]->feature.rect[2].p0 ) if( nodes[4]->feature.rect[2].p0 )
tmp[4] = calc_sum(nodes[4]->feature.rect[2],p_offset) * nodes[4]->feature.rect[2].weight; tmp[4] = calc_sum(nodes[4]->feature.rect[2],p_offset) * nodes[4]->feature.rect[2].weight;
if( nodes[5]->feature.rect[2].p0 ) if( nodes[5]->feature.rect[2].p0 )
tmp[5] = calc_sum(nodes[5]->feature.rect[2],p_offset) * nodes[5]->feature.rect[2].weight; tmp[5] = calc_sum(nodes[5]->feature.rect[2],p_offset) * nodes[5]->feature.rect[2].weight;
if( nodes[6]->feature.rect[2].p0 ) if( nodes[6]->feature.rect[2].p0 )
tmp[6] = calc_sum(nodes[6]->feature.rect[2],p_offset) * nodes[6]->feature.rect[2].weight; tmp[6] = calc_sum(nodes[6]->feature.rect[2],p_offset) * nodes[6]->feature.rect[2].weight;
if( nodes[7]->feature.rect[2].p0 ) if( nodes[7]->feature.rect[2].p0 )
tmp[7] = calc_sum(nodes[7]->feature.rect[2],p_offset) * nodes[7]->feature.rect[2].weight; tmp[7] = calc_sum(nodes[7]->feature.rect[2],p_offset) * nodes[7]->feature.rect[2].weight;
sum = _mm256_add_ps(sum, _mm256_load_ps(tmp)); sum = _mm256_add_ps(sum, _mm256_load_ps(tmp));
__m256 alpha0 = _mm256_set_ps(classifiers[7]->alpha[0],classifiers[6]->alpha[0],classifiers[5]->alpha[0],classifiers[4]->alpha[0],classifiers[3]->alpha[0], __m256 alpha0 = _mm256_set_ps(classifiers[7]->alpha[0],classifiers[6]->alpha[0],classifiers[5]->alpha[0],classifiers[4]->alpha[0],classifiers[3]->alpha[0],
classifiers[2]->alpha[0],classifiers[1]->alpha[0],classifiers[0]->alpha[0]); classifiers[2]->alpha[0],classifiers[1]->alpha[0],classifiers[0]->alpha[0]);
__m256 alpha1 = _mm256_set_ps(classifiers[7]->alpha[1],classifiers[6]->alpha[1],classifiers[5]->alpha[1],classifiers[4]->alpha[1],classifiers[3]->alpha[1], __m256 alpha1 = _mm256_set_ps(classifiers[7]->alpha[1],classifiers[6]->alpha[1],classifiers[5]->alpha[1],classifiers[4]->alpha[1],classifiers[3]->alpha[1],
classifiers[2]->alpha[1],classifiers[1]->alpha[1],classifiers[0]->alpha[1]); classifiers[2]->alpha[1],classifiers[1]->alpha[1],classifiers[0]->alpha[1]);
__m256 outBuf = _mm256_blendv_ps(alpha0, alpha1, _mm256_cmp_ps(t, sum, _CMP_LE_OQ )); __m256 outBuf = _mm256_blendv_ps(alpha0, alpha1, _mm256_cmp_ps(t, sum, _CMP_LE_OQ ));
outBuf = _mm256_hadd_ps(outBuf, outBuf); outBuf = _mm256_hadd_ps(outBuf, outBuf);
outBuf = _mm256_hadd_ps(outBuf, outBuf); outBuf = _mm256_hadd_ps(outBuf, outBuf);
_mm256_store_ps(buf, outBuf); _mm256_store_ps(buf, outBuf);
stage_sum+=(buf[0]+buf[4]);//(buf[0]+buf[1]+buf[2]+buf[3]+buf[4]+buf[5]+buf[6]+buf[7]); stage_sum+=(buf[0]+buf[4]);//(buf[0]+buf[1]+buf[2]+buf[3]+buf[4]+buf[5]+buf[6]+buf[7]);
} }
for( ; j < cascade->stage_classifier[i].count; j++ ) for( ; j < cascade->stage_classifier[i].count; j++ )
{ {
CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j; CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
CvHidHaarTreeNode* node = classifier->node; CvHidHaarTreeNode* node = classifier->node;
double t = node->threshold*variance_norm_factor; double t = node->threshold*variance_norm_factor;
double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
if( node->feature.rect[2].p0 ) if( node->feature.rect[2].p0 )
sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight; sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
stage_sum += classifier->alpha[sum >= t]; stage_sum += classifier->alpha[sum >= t];
} }
} }
if( stage_sum < cascade->stage_classifier[i].threshold ) if( stage_sum < cascade->stage_classifier[i].threshold )
return -i; return -i;
} }
} }
else else
#endif #endif
#ifdef CV_HAAR_USE_SSE && !CV_HAAR_USE_AVX //old SSE optimization #if defined CV_HAAR_USE_SSE && CV_HAAR_USE_SSE && !CV_HAAR_USE_AVX //old SSE optimization
if(haveSSE2) if(haveSSE2)
{ {
for( i = start_stage; i < cascade->count; i++ ) for( i = start_stage; i < cascade->count; i++ )
{ {
__m128d stage_sum = _mm_setzero_pd(); __m128d stage_sum = _mm_setzero_pd();
if( cascade->stage_classifier[i].two_rects ) if( cascade->stage_classifier[i].two_rects )
{ {
for( j = 0; j < cascade->stage_classifier[i].count; j++ ) for( j = 0; j < cascade->stage_classifier[i].count; j++ )
{ {
CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j; CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
CvHidHaarTreeNode* node = classifier->node; CvHidHaarTreeNode* node = classifier->node;
// ayasin - NHM perf optim. Avoid use of costly flaky jcc // ayasin - NHM perf optim. Avoid use of costly flaky jcc
__m128d t = _mm_set_sd(node->threshold*variance_norm_factor); __m128d t = _mm_set_sd(node->threshold*variance_norm_factor);
__m128d a = _mm_set_sd(classifier->alpha[0]); __m128d a = _mm_set_sd(classifier->alpha[0]);
__m128d b = _mm_set_sd(classifier->alpha[1]); __m128d b = _mm_set_sd(classifier->alpha[1]);
__m128d sum = _mm_set_sd(calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight + __m128d sum = _mm_set_sd(calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight +
calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight); calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight);
t = _mm_cmpgt_sd(t, sum); t = _mm_cmpgt_sd(t, sum);
stage_sum = _mm_add_sd(stage_sum, _mm_blendv_pd(b, a, t)); stage_sum = _mm_add_sd(stage_sum, _mm_blendv_pd(b, a, t));
} }
} }
else else
{ {
for( j = 0; j < cascade->stage_classifier[i].count; j++ ) for( j = 0; j < cascade->stage_classifier[i].count; j++ )
{ {
CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j; CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
CvHidHaarTreeNode* node = classifier->node; CvHidHaarTreeNode* node = classifier->node;
// ayasin - NHM perf optim. Avoid use of costly flaky jcc // ayasin - NHM perf optim. Avoid use of costly flaky jcc
__m128d t = _mm_set_sd(node->threshold*variance_norm_factor); __m128d t = _mm_set_sd(node->threshold*variance_norm_factor);
__m128d a = _mm_set_sd(classifier->alpha[0]); __m128d a = _mm_set_sd(classifier->alpha[0]);
__m128d b = _mm_set_sd(classifier->alpha[1]); __m128d b = _mm_set_sd(classifier->alpha[1]);
double _sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double _sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
_sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; _sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
if( node->feature.rect[2].p0 ) if( node->feature.rect[2].p0 )
_sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight; _sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
__m128d sum = _mm_set_sd(_sum); __m128d sum = _mm_set_sd(_sum);
t = _mm_cmpgt_sd(t, sum); t = _mm_cmpgt_sd(t, sum);
stage_sum = _mm_add_sd(stage_sum, _mm_blendv_pd(b, a, t)); stage_sum = _mm_add_sd(stage_sum, _mm_blendv_pd(b, a, t));
} }
} }
__m128d i_threshold = _mm_set1_pd(cascade->stage_classifier[i].threshold); __m128d i_threshold = _mm_set1_pd(cascade->stage_classifier[i].threshold);
if( _mm_comilt_sd(stage_sum, i_threshold) ) if( _mm_comilt_sd(stage_sum, i_threshold) )
return -i; return -i;
} }
} }
else else
#endif #endif
{ {
for( i = start_stage; i < cascade->count; i++ ) for( i = start_stage; i < cascade->count; i++ )
{ {
stage_sum = 0.0; stage_sum = 0.0;
if( cascade->stage_classifier[i].two_rects ) if( cascade->stage_classifier[i].two_rects )
{ {
for( j = 0; j < cascade->stage_classifier[i].count; j++ ) for( j = 0; j < cascade->stage_classifier[i].count; j++ )
{ {
CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j; CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
CvHidHaarTreeNode* node = classifier->node; CvHidHaarTreeNode* node = classifier->node;
double t = node->threshold*variance_norm_factor; double t = node->threshold*variance_norm_factor;
double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
stage_sum += classifier->alpha[sum >= t]; stage_sum += classifier->alpha[sum >= t];
} }
} }
else else
{ {
for( j = 0; j < cascade->stage_classifier[i].count; j++ ) for( j = 0; j < cascade->stage_classifier[i].count; j++ )
{ {
CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j; CvHidHaarClassifier* classifier = cascade->stage_classifier[i].classifier + j;
CvHidHaarTreeNode* node = classifier->node; CvHidHaarTreeNode* node = classifier->node;
double t = node->threshold*variance_norm_factor; double t = node->threshold*variance_norm_factor;
double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight; double sum = calc_sum(node->feature.rect[0],p_offset) * node->feature.rect[0].weight;
sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight; sum += calc_sum(node->feature.rect[1],p_offset) * node->feature.rect[1].weight;
if( node->feature.rect[2].p0 ) if( node->feature.rect[2].p0 )
sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight; sum += calc_sum(node->feature.rect[2],p_offset) * node->feature.rect[2].weight;
stage_sum += classifier->alpha[sum >= t]; stage_sum += classifier->alpha[sum >= t];
} }
} }
if( stage_sum < cascade->stage_classifier[i].threshold ) if( stage_sum < cascade->stage_classifier[i].threshold )
return -i; return -i;
} }
} }
} }
else else
{ {
for( i = start_stage; i < cascade->count; i++ ) for( i = start_stage; i < cascade->count; i++ )
{ {
stage_sum = 0.0; stage_sum = 0.0;
int j = 0; int k = 0;
#ifdef CV_HAAR_USE_AVX #ifdef CV_HAAR_USE_AVX
if(haveAVX) if(haveAVX)
{ {
for( ; j < cascade->stage_classifier[i].count-8; j+=8 ) for( ; k < cascade->stage_classifier[i].count-8; k+=8 )
{ {
stage_sum += icvEvalHidHaarClassifierAVX( stage_sum += icvEvalHidHaarClassifierAVX(
cascade->stage_classifier[i].classifier+j, cascade->stage_classifier[i].classifier+k,
variance_norm_factor, p_offset ); variance_norm_factor, p_offset );
} }
} }
#endif #endif
for(; j < cascade->stage_classifier[i].count; j++ ) for(; k < cascade->stage_classifier[i].count; k++ )
{ {
stage_sum += icvEvalHidHaarClassifier( stage_sum += icvEvalHidHaarClassifier(
cascade->stage_classifier[i].classifier + j, cascade->stage_classifier[i].classifier + k,
variance_norm_factor, p_offset ); variance_norm_factor, p_offset );
} }
if( stage_sum < cascade->stage_classifier[i].threshold ) if( stage_sum < cascade->stage_classifier[i].threshold )
return -i; return -i;
} }
} }
//_mm256_zeroupper(); //_mm256_zeroupper();
return 1; return 1;
} }

8
samples/cpp/pca.cpp
View File

@ -50,7 +50,7 @@ using namespace std;
/////////////////////// ///////////////////////
// Functions // Functions
void read_imgList(const string& filename, vector<Mat>& images) { static void read_imgList(const string& filename, vector<Mat>& images) {
std::ifstream file(filename.c_str(), ifstream::in); std::ifstream file(filename.c_str(), ifstream::in);
if (!file) { if (!file) {
string error_message = "No valid input file was given, please check the given filename."; string error_message = "No valid input file was given, please check the given filename.";
@ -62,7 +62,7 @@ void read_imgList(const string& filename, vector<Mat>& images) {
} }
} }
Mat formatImagesForPCA(const vector<Mat> &data) static Mat formatImagesForPCA(const vector<Mat> &data)
{ {
Mat dst(data.size(), data[0].rows*data[0].cols, CV_32F); Mat dst(data.size(), data[0].rows*data[0].cols, CV_32F);
for(unsigned int i = 0; i < data.size(); i++) for(unsigned int i = 0; i < data.size(); i++)
@ -74,7 +74,7 @@ Mat formatImagesForPCA(const vector<Mat> &data)
return dst; return dst;
} }
Mat toGrayscale(InputArray _src) { static Mat toGrayscale(InputArray _src) {
Mat src = _src.getMat(); Mat src = _src.getMat();
// only allow one channel // only allow one channel
if(src.channels() != 1) { if(src.channels() != 1) {
@ -95,7 +95,7 @@ struct params
string winName; string winName;
}; };
void onTrackbar(int pos, void* ptr) static void onTrackbar(int pos, void* ptr)
{ {
cout << "Retained Variance = " << pos << "% "; cout << "Retained Variance = " << pos << "% ";
cout << "re-calculating PCA..." << std::flush; cout << "re-calculating PCA..." << std::flush;