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

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

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

275
modules/gpu/src/cascadeclassifier.cpp
View File

@@ -41,6 +41,8 @@
//M*/
#include "precomp.hpp"
#include <vector>
#include <iostream>
#include "opencv2/objdetect/objdetect_c.h"
using namespace cv;
@@ -73,37 +75,6 @@ public:
virtual bool read(const String& classifierAsXml) = 0;
};
#ifndef HAVE_OPENCV_GPULEGACY
struct cv::gpu::CascadeClassifier_GPU::HaarCascade : cv::gpu::CascadeClassifier_GPU::CascadeClassifierImpl
{
public:
HaarCascade()
{
throw_no_cuda();
}
unsigned int process(const GpuMat&, GpuMat&, float, int, bool, bool, cv::Size, cv::Size)
{
throw_no_cuda();
return 0;
}
cv::Size getClassifierCvSize() const
{
throw_no_cuda();
return cv::Size();
}
bool read(const String&)
{
throw_no_cuda();
return false;
}
};
#else
struct cv::gpu::CascadeClassifier_GPU::HaarCascade : cv::gpu::CascadeClassifier_GPU::CascadeClassifierImpl
{
public:
@@ -313,8 +284,6 @@ private:
virtual ~HaarCascade(){}
};
#endif
cv::Size operator -(const cv::Size& a, const cv::Size& b)
{
return cv::Size(a.width - b.width, a.height - b.height);
@@ -508,8 +477,6 @@ private:
resuzeBuffer.create(frame, CV_8UC1);
integral.create(frame.height + 1, integralFactor * (frame.width + 1), CV_32SC1);
#ifdef HAVE_OPENCV_GPULEGACY
NcvSize32u roiSize;
roiSize.width = frame.width;
roiSize.height = frame.height;
@@ -520,7 +487,6 @@ private:
Ncv32u bufSize;
ncvSafeCall( nppiStIntegralGetSize_8u32u(roiSize, &bufSize, prop) );
integralBuffer.create(1, bufSize, CV_8UC1);
#endif
candidates.create(1 , frame.width >> 1, CV_32SC4);
}
@@ -756,3 +722,240 @@ bool cv::gpu::CascadeClassifier_GPU::load(const String& filename)
}
#endif
//////////////////////////////////////////////////////////////////////////////////////////////////////
#if defined (HAVE_CUDA)
struct RectConvert
{
Rect operator()(const NcvRect32u& nr) const { return Rect(nr.x, nr.y, nr.width, nr.height); }
NcvRect32u operator()(const Rect& nr) const
{
NcvRect32u rect;
rect.x = nr.x;
rect.y = nr.y;
rect.width = nr.width;
rect.height = nr.height;
return rect;
}
};
void groupRectangles(std::vector<NcvRect32u> &hypotheses, int groupThreshold, double eps, std::vector<Ncv32u> *weights)
{
std::vector<Rect> rects(hypotheses.size());
std::transform(hypotheses.begin(), hypotheses.end(), rects.begin(), RectConvert());
if (weights)
{
std::vector<int> weights_int;
weights_int.assign(weights->begin(), weights->end());
cv::groupRectangles(rects, weights_int, groupThreshold, eps);
}
else
{
cv::groupRectangles(rects, groupThreshold, eps);
}
std::transform(rects.begin(), rects.end(), hypotheses.begin(), RectConvert());
hypotheses.resize(rects.size());
}
NCVStatus loadFromXML(const String &filename,
HaarClassifierCascadeDescriptor &haar,
std::vector<HaarStage64> &haarStages,
std::vector<HaarClassifierNode128> &haarClassifierNodes,
std::vector<HaarFeature64> &haarFeatures)
{
NCVStatus ncvStat;
haar.NumStages = 0;
haar.NumClassifierRootNodes = 0;
haar.NumClassifierTotalNodes = 0;
haar.NumFeatures = 0;
haar.ClassifierSize.width = 0;
haar.ClassifierSize.height = 0;
haar.bHasStumpsOnly = true;
haar.bNeedsTiltedII = false;
Ncv32u curMaxTreeDepth;
std::vector<char> xmlFileCont;
std::vector<HaarClassifierNode128> h_TmpClassifierNotRootNodes;
haarStages.resize(0);
haarClassifierNodes.resize(0);
haarFeatures.resize(0);
Ptr<CvHaarClassifierCascade> oldCascade = (CvHaarClassifierCascade*)cvLoad(filename.c_str(), 0, 0, 0);
if (oldCascade.empty())
{
return NCV_HAAR_XML_LOADING_EXCEPTION;
}
haar.ClassifierSize.width = oldCascade->orig_window_size.width;
haar.ClassifierSize.height = oldCascade->orig_window_size.height;
int stagesCound = oldCascade->count;
for(int s = 0; s < stagesCound; ++s) // by stages
{
HaarStage64 curStage;
curStage.setStartClassifierRootNodeOffset(static_cast<Ncv32u>(haarClassifierNodes.size()));
curStage.setStageThreshold(oldCascade->stage_classifier[s].threshold);
int treesCount = oldCascade->stage_classifier[s].count;
for(int t = 0; t < treesCount; ++t) // by trees
{
Ncv32u nodeId = 0;
CvHaarClassifier* tree = &oldCascade->stage_classifier[s].classifier[t];
int nodesCount = tree->count;
for(int n = 0; n < nodesCount; ++n) //by features
{
CvHaarFeature* feature = &tree->haar_feature[n];
HaarClassifierNode128 curNode;
curNode.setThreshold(tree->threshold[n]);
NcvBool bIsLeftNodeLeaf = false;
NcvBool bIsRightNodeLeaf = false;
HaarClassifierNodeDescriptor32 nodeLeft;
if ( tree->left[n] <= 0 )
{
Ncv32f leftVal = tree->alpha[-tree->left[n]];
ncvStat = nodeLeft.create(leftVal);
ncvAssertReturn(ncvStat == NCV_SUCCESS, ncvStat);
bIsLeftNodeLeaf = true;
}
else
{
Ncv32u leftNodeOffset = tree->left[n];
nodeLeft.create((Ncv32u)(h_TmpClassifierNotRootNodes.size() + leftNodeOffset - 1));
haar.bHasStumpsOnly = false;
}
curNode.setLeftNodeDesc(nodeLeft);
HaarClassifierNodeDescriptor32 nodeRight;
if ( tree->right[n] <= 0 )
{
Ncv32f rightVal = tree->alpha[-tree->right[n]];
ncvStat = nodeRight.create(rightVal);
ncvAssertReturn(ncvStat == NCV_SUCCESS, ncvStat);
bIsRightNodeLeaf = true;
}
else
{
Ncv32u rightNodeOffset = tree->right[n];
nodeRight.create((Ncv32u)(h_TmpClassifierNotRootNodes.size() + rightNodeOffset - 1));
haar.bHasStumpsOnly = false;
}
curNode.setRightNodeDesc(nodeRight);
Ncv32u tiltedVal = feature->tilted;
haar.bNeedsTiltedII = (tiltedVal != 0);
Ncv32u featureId = 0;
for(int l = 0; l < CV_HAAR_FEATURE_MAX; ++l) //by rects
{
Ncv32u rectX = feature->rect[l].r.x;
Ncv32u rectY = feature->rect[l].r.y;
Ncv32u rectWidth = feature->rect[l].r.width;
Ncv32u rectHeight = feature->rect[l].r.height;
Ncv32f rectWeight = feature->rect[l].weight;
if (rectWeight == 0/* && rectX == 0 &&rectY == 0 && rectWidth == 0 && rectHeight == 0*/)
break;
HaarFeature64 curFeature;
ncvStat = curFeature.setRect(rectX, rectY, rectWidth, rectHeight, haar.ClassifierSize.width, haar.ClassifierSize.height);
curFeature.setWeight(rectWeight);
ncvAssertReturn(NCV_SUCCESS == ncvStat, ncvStat);
haarFeatures.push_back(curFeature);
featureId++;
}
HaarFeatureDescriptor32 tmpFeatureDesc;
ncvStat = tmpFeatureDesc.create(haar.bNeedsTiltedII, bIsLeftNodeLeaf, bIsRightNodeLeaf,
featureId, static_cast<Ncv32u>(haarFeatures.size()) - featureId);
ncvAssertReturn(NCV_SUCCESS == ncvStat, ncvStat);
curNode.setFeatureDesc(tmpFeatureDesc);
if (!nodeId)
{
//root node
haarClassifierNodes.push_back(curNode);
curMaxTreeDepth = 1;
}
else
{
//other node
h_TmpClassifierNotRootNodes.push_back(curNode);
curMaxTreeDepth++;
}
nodeId++;
}
}
curStage.setNumClassifierRootNodes(treesCount);
haarStages.push_back(curStage);
}
//fill in cascade stats
haar.NumStages = static_cast<Ncv32u>(haarStages.size());
haar.NumClassifierRootNodes = static_cast<Ncv32u>(haarClassifierNodes.size());
haar.NumClassifierTotalNodes = static_cast<Ncv32u>(haar.NumClassifierRootNodes + h_TmpClassifierNotRootNodes.size());
haar.NumFeatures = static_cast<Ncv32u>(haarFeatures.size());
//merge root and leaf nodes in one classifiers array
Ncv32u offsetRoot = static_cast<Ncv32u>(haarClassifierNodes.size());
for (Ncv32u i=0; i<haarClassifierNodes.size(); i++)
{
HaarFeatureDescriptor32 featureDesc = haarClassifierNodes[i].getFeatureDesc();
HaarClassifierNodeDescriptor32 nodeLeft = haarClassifierNodes[i].getLeftNodeDesc();
if (!featureDesc.isLeftNodeLeaf())
{
Ncv32u newOffset = nodeLeft.getNextNodeOffset() + offsetRoot;
nodeLeft.create(newOffset);
}
haarClassifierNodes[i].setLeftNodeDesc(nodeLeft);
HaarClassifierNodeDescriptor32 nodeRight = haarClassifierNodes[i].getRightNodeDesc();
if (!featureDesc.isRightNodeLeaf())
{
Ncv32u newOffset = nodeRight.getNextNodeOffset() + offsetRoot;
nodeRight.create(newOffset);
}
haarClassifierNodes[i].setRightNodeDesc(nodeRight);
}
for (Ncv32u i=0; i<h_TmpClassifierNotRootNodes.size(); i++)
{
HaarFeatureDescriptor32 featureDesc = h_TmpClassifierNotRootNodes[i].getFeatureDesc();
HaarClassifierNodeDescriptor32 nodeLeft = h_TmpClassifierNotRootNodes[i].getLeftNodeDesc();
if (!featureDesc.isLeftNodeLeaf())
{
Ncv32u newOffset = nodeLeft.getNextNodeOffset() + offsetRoot;
nodeLeft.create(newOffset);
}
h_TmpClassifierNotRootNodes[i].setLeftNodeDesc(nodeLeft);
HaarClassifierNodeDescriptor32 nodeRight = h_TmpClassifierNotRootNodes[i].getRightNodeDesc();
if (!featureDesc.isRightNodeLeaf())
{
Ncv32u newOffset = nodeRight.getNextNodeOffset() + offsetRoot;
nodeRight.create(newOffset);
}
h_TmpClassifierNotRootNodes[i].setRightNodeDesc(nodeRight);
haarClassifierNodes.push_back(h_TmpClassifierNotRootNodes[i]);
}
return NCV_SUCCESS;
}
#endif /* HAVE_CUDA */