added getDescriptors support into gpu HOG, also added commented test for this feature

modules/gpu/include/opencv2/gpu/gpu.hpp

@@ -1001,14 +1001,13 @@ namespace cv
             void setSVMDetector(const vector<float>& detector);
             bool checkDetectorSize() const;
 
+            void computeBlockHistograms(const GpuMat& img);
             void detect(const GpuMat& img, vector<Point>& found_locations, double hit_threshold=0, 
                         Size win_stride=Size(), Size padding=Size());
             void detectMultiScale(const GpuMat& img, vector<Rect>& found_locations, 
                                   double hit_threshold=0, Size win_stride=Size(), Size padding=Size(),
                                   double scale0=1.05, int group_threshold=2);
-
+            void getDescriptors(const GpuMat& img, Size win_stride, GpuMat& descriptors);
-            ////TODO: test it
-            //void getDescriptors(const GpuMat& img, Size win_stride, vector<GpuMat>& descriptors)
 
             Size win_size;
             Size block_size;
@@ -1035,7 +1034,6 @@ namespace cv
             static int numPartsWithin(int size, int part_size, int stride);
             static Size numPartsWithin(Size size, Size part_size, Size stride);
 
-            void computeBlockHistograms(const GpuMat& img);            
             void computeGradient(const GpuMat& img, GpuMat& grad, GpuMat& qangle);
 
             GpuMat grad, qangle;

modules/gpu/src/cuda/hog.cu

@@ -397,11 +397,9 @@ __global__ void classify_hists_kernel_many_blocks(const int img_win_width, const
 }
 
 
-// We only support win_stride_x == block_stride_x, win_stride_y == block_stride_y
 void classify_hists(int win_height, int win_width, int block_stride_y, int block_stride_x, 
-                    int win_stride_y, int win_stride_x,
+                    int win_stride_y, int win_stride_x, int height, int width, float* block_hists, 
-                    int height, int width, float* block_hists, float* coefs, 
+                    float* coefs, float free_coef, float threshold, unsigned char* labels)
-                    float free_coef, float threshold, unsigned char* labels)
 {   
     const int nthreads = 256;
     const int nblocks = 1;
@@ -425,8 +423,54 @@ void classify_hists(int win_height, int win_width, int block_stride_y, int block
     cudaSafeCall(cudaThreadSynchronize());
 }
 
+//----------------------------------------------------------------------------
+// Extract descriptors
 
-//------------------------------------------------------------
+
+template <int nthreads>
+__global__ void extract_descriptors_kernel(const int img_win_width, const int img_block_width, 
+                                           const int win_block_stride_x, const int win_block_stride_y,
+                                           const float* block_hists, PtrElemStepf descriptors)
+{
+    // Get left top corner of the window in src
+    const float* hist = block_hists + (blockIdx.y * win_block_stride_y * img_block_width + 
+                                       blockIdx.x * win_block_stride_x) * cblock_hist_size;
+
+    // Get left top corner of the window in dst
+    float* descriptor = descriptors.ptr(blockIdx.y * gridDim.x + blockIdx.x);
+
+    // Copy elements from src to dst
+    for (int i = threadIdx.x; i < cdescr_size; i += nthreads)
+    {
+        int offset_y = i / cdescr_width;
+        int offset_x = i - offset_y * cdescr_width;
+        descriptor[i] = hist[offset_y * img_block_width * cblock_hist_size + offset_x];
+    }
+}
+
+
+void extract_descriptors(int win_height, int win_width, int block_stride_y, int block_stride_x, 
+                         int win_stride_y, int win_stride_x, int height, int width, float* block_hists, 
+                         DevMem2Df descriptors)
+{
+    const int nthreads = 256;
+
+    int win_block_stride_x = win_stride_x / block_stride_x;
+    int win_block_stride_y = win_stride_y / block_stride_y;
+    int img_win_width = (width - win_width + win_stride_x) / win_stride_x;
+    int img_win_height = (height - win_height + win_stride_y) / win_stride_y;
+    dim3 threads(nthreads, 1);
+    dim3 grid(img_win_width, img_win_height);
+
+    int img_block_width = (width - CELLS_PER_BLOCK_X * CELL_WIDTH + block_stride_x) / 
+                          block_stride_x;
+    extract_descriptors_kernel<nthreads><<<grid, threads>>>(
+        img_win_width, img_block_width, win_block_stride_x, win_block_stride_y, 
+        block_hists, descriptors);
+    cudaSafeCall(cudaThreadSynchronize());
+}
+
+//----------------------------------------------------------------------------
 // Gradients computation
 
 
@@ -481,7 +525,7 @@ __global__ void compute_gradients_8UC4_kernel(int height, int width, const PtrEl
 
         float3 dx = make_float3(sqrtf(b.x) - sqrtf(a.x), 
                                 sqrtf(b.y) - sqrtf(a.y), 
-                                sqrtf(b.z) - sqrtf(a.z));        
+                                sqrtf(b.z) - sqrtf(a.z));    
         float3 dy = make_float3(0.f, 0.f, 0.f);
 
         if (blockIdx.y > 0 && blockIdx.y < height - 1)

modules/gpu/src/hog.cpp

@@ -51,8 +51,10 @@ double cv::gpu::HOGDescriptor::getWinSigma() const { throw_nogpu(); return 0; }
 bool cv::gpu::HOGDescriptor::checkDetectorSize() const { throw_nogpu(); return false; }
 void cv::gpu::HOGDescriptor::setSVMDetector(const vector<float>&) { throw_nogpu(); }
 void cv::gpu::HOGDescriptor::computeGradient(const GpuMat&, GpuMat&, GpuMat&) { throw_nogpu(); }
+void cv::gpu::HOGDescriptor::computeBlockHistograms(const GpuMat&) { throw_nogpu(); }
 void cv::gpu::HOGDescriptor::detect(const GpuMat&, vector<Point>&, double, Size, Size) { throw_nogpu(); }
 void cv::gpu::HOGDescriptor::detectMultiScale(const GpuMat&, vector<Rect>&, double, Size, Size, double, int) { throw_nogpu(); }
+void cv::gpu::HOGDescriptor::getDescriptors(const GpuMat&, Size, GpuMat&) { throw_nogpu(); }
 std::vector<float> cv::gpu::HOGDescriptor::getDefaultPeopleDetector() { throw_nogpu(); return std::vector<float>(); }
 std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_48x96() { throw_nogpu(); return std::vector<float>(); }
 std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_64x128() { throw_nogpu(); return std::vector<float>(); }
@@ -76,6 +78,10 @@ void classify_hists(int win_height, int win_width, int block_stride_y,
                     int width, float* block_hists, float* coefs, float free_coef, 
                     float threshold, unsigned char* labels);
 
+void extract_descriptors(int win_height, int win_width, int block_stride_y, int block_stride_x, 
+                         int win_stride_y, int win_stride_x, int height, int width, float* block_hists, 
+                         cv::gpu::DevMem2Df descriptors);
+
 void compute_gradients_8UC1(int nbins, int height, int width, const cv::gpu::DevMem2D& img, 
                             float angle_scale, cv::gpu::DevMem2Df grad, cv::gpu::DevMem2D qangle);
 void compute_gradients_8UC4(int nbins, int height, int width, const cv::gpu::DevMem2D& img, 
@@ -212,39 +218,23 @@ void cv::gpu::HOGDescriptor::computeBlockHistograms(const GpuMat& img)
 }
 
 
-////TODO: test it
+void cv::gpu::HOGDescriptor::getDescriptors(const GpuMat& img, Size win_stride, GpuMat& descriptors)
-//void cv::gpu::HOGDescriptor::getDescriptors(const GpuMat& img, Size win_stride, 
+{
-//                                            vector<GpuMat>& descriptors)
+    CV_Assert(win_stride.width % block_stride.width == 0 &&
-//{
+              win_stride.height % block_stride.height == 0);
-//    CV_Assert(win_stride.width % block_stride.width == 0 &&
+
-//              win_stride.height % block_stride.height == 0);
+    computeBlockHistograms(img);
-//
+
-//    computeBlockHistograms(img);
+    const int block_hist_size = getBlockHistogramSize();
-//
+    Size blocks_per_win = numPartsWithin(win_size, block_size, block_stride);
-//    Size blocks_per_img = numPartsWithin(img.size(), block_size, block_stride);
+    Size wins_per_img = numPartsWithin(img.size(), win_size, win_stride);
-//    GpuMat hists_reshaped = block_hists.reshape(0, blocks_per_img.height);
+
-//
+    descriptors.create(wins_per_img.area(), blocks_per_win.area() * block_hist_size, CV_32F);
-//    const int block_hist_size = getBlockHistogramSize();
+
-//    Size blocks_per_win = numPartsWithin(win_size, block_size, block_stride);
+    hog::extract_descriptors(win_size.height, win_size.width, block_stride.height, block_stride.width, 
-//    Size wins_per_img = numPartsWithin(img.size(), win_size, win_stride);
+                             win_stride.height, win_stride.width, img.rows, img.cols, block_hists.ptr<float>(), 
-//
+                             descriptors);
-//    descriptors.resize(wins_per_img.area());
+}
-//    for (int i = 0; i < wins_per_img.height; ++i)
-//    {
-//        for (int j = 0; j < wins_per_img.width; ++j)
-//        {
-//            Range rows;
-//            rows.start = i * (blocks_per_win.height + 1);
-//            rows.end = rows.start + blocks_per_win.height;
-//
-//            Range cols;
-//            cols.start = j * (blocks_per_win.width + 1) * block_hist_size;
-//            cols.end = cols.start + blocks_per_win.width * block_hist_size;
-//
-//            descriptors[i * wins_per_img.width + j] = hists_reshaped(rows, cols);
-//        }
-//    }
-//}
 
 
 void cv::gpu::HOGDescriptor::detect(const GpuMat& img, vector<Point>& hits, double hit_threshold, 

samples/gpu/gpu_hog.cpp

@@ -225,7 +225,11 @@ void App::RunOpencvGui()
             vc >> frame;
         }
         else
+        {
             frame = imread(settings.src);
+            if (frame.empty())
+                throw exception(string("Can't open image file: " + settings.src).c_str());
+        }
 
         Mat img_aux, img, img_to_show;
         gpu::GpuMat gpu_img;

tests/gpu/src/hog.cpp

@@ -51,9 +51,43 @@ using namespace std;
     ts->set_failed_test_info(err); \
     return; }
 
-struct CV_GpuHogTest : public CvTest 
+struct CV_GpuHogDetectionTest: public CvTest 
 {
-    CV_GpuHogTest() : CvTest( "GPU-HOG", "HOGDescriptor" ) {}
+    CV_GpuHogDetectionTest(): CvTest( "GPU-HOG-detect", "HOGDescriptorDetection" ) {}
+
+    void run(int) 
+    {       
+        try 
+        {
+            cv::Mat img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/road.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+
+#ifdef DUMP
+            f.open((std::string(ts->get_data_path()) + "hog/expected_output.bin").c_str(), std::ios_base::binary);
+            CHECK(f.is_open(), CvTS::FAIL_GENERIC);          
+#else
+            f.open((std::string(ts->get_data_path()) + "hog/expected_output.bin").c_str(), std::ios_base::binary);
+            CHECK(f.is_open(), CvTS::FAIL_MISSING_TEST_DATA);          
+#endif
+            
+            // Test on color image
+            cv::Mat img;
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            test(img);
+
+            // Test on gray image
+            cv::cvtColor(img_rgb, img, CV_BGR2GRAY);
+            test(img);
+
+            f.close();
+        }
+        catch (const cv::Exception& e)
+        {
+            f.close();
+            if (!check_and_treat_gpu_exception(e, ts)) throw;
+            return;
+        }
+    }
 
 #ifdef DUMP
     void dump(const cv::Mat& block_hists, const std::vector<cv::Point>& locations) 
@@ -168,45 +202,115 @@ struct CV_GpuHogTest : public CvTest
 #endif
     }
 
-
-    void run(int) 
-    {       
-        try 
-        {
-            cv::Mat img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/road.png");
-            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
-
-#ifdef DUMP
-            f.open((std::string(ts->get_data_path()) + "hog/expected_output.bin").c_str(), std::ios_base::binary);
-            CHECK(f.is_open(), CvTS::FAIL_GENERIC);          
-#else
-            f.open((std::string(ts->get_data_path()) + "hog/expected_output.bin").c_str(), std::ios_base::binary);
-            CHECK(f.is_open(), CvTS::FAIL_MISSING_TEST_DATA);          
-#endif
-            
-            // Test on color image
-            cv::Mat img;
-            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
-            test(img);
-
-            // Test on gray image
-            cv::cvtColor(img_rgb, img, CV_BGR2GRAY);
-            test(img);
-
-            f.close();
-        }
-        catch (const cv::Exception& e)
-        {
-            f.close();
-            if (!check_and_treat_gpu_exception(e, ts)) throw;
-            return;
-        }
-    }
-
 #ifdef DUMP
     std::ofstream f;
 #else
     std::ifstream f;
 #endif
 
-} gpu_hog_test;
+} gpu_hog_detection_test;
+
+
+struct CV_GpuHogGetDescriptorsTest: public CvTest 
+{
+    CV_GpuHogGetDescriptorsTest(): CvTest("GPU-HOG-getDescriptors", "HOGDescriptorGetDescriptors") {}
+
+    void run(int)
+    {
+        try 
+        {
+            // Load image (e.g. train data, composed from windows)
+            cv::Mat img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/train_data.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+
+            // Convert to C4
+            cv::Mat img;
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            cv::gpu::GpuMat d_img(img);
+
+            cv::Size win_size(64, 128);
+            cv::gpu::HOGDescriptor hog(win_size);
+
+            // Convert train images into feature vectors (train table)
+            cv::gpu::GpuMat descriptors;
+            hog.getDescriptors(d_img, win_size, descriptors);
+
+            // Check size of the result train table
+            wins_per_img_x = 3;
+            wins_per_img_y = 2;
+            blocks_per_win_x = 7;
+            blocks_per_win_y = 15;
+            block_hist_size = 36;
+            cv::Size descr_size_expected = cv::Size(blocks_per_win_x * blocks_per_win_y * block_hist_size, 
+                                                    wins_per_img_x * wins_per_img_y);                                                
+            CHECK(descriptors.size() == descr_size_expected, CvTS::FAIL_INVALID_OUTPUT);
+
+            /* Now we want to extract the same feature vectors, but from single images. NOTE: results will 
+            be defferent, due to border values interpolation. Using of many small images is slower, however we 
+            wont't call getDescriptors and will use computeBlockHistograms instead of. computeBlockHistograms 
+            works good, it can be checked in the gpu_hog sample */
+
+            img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/positive1.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            hog.computeBlockHistograms(cv::gpu::GpuMat(img));
+            // Everything is fine with interpolation for left top subimage
+            CHECK(cv::norm(hog.block_hists, descriptors.rowRange(0, 1)) == 0.f, CvTS::FAIL_INVALID_OUTPUT);
+
+            img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/positive2.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            hog.computeBlockHistograms(cv::gpu::GpuMat(img));
+            compare_inner_parts(hog.block_hists, descriptors.rowRange(1, 2));
+
+            img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/negative1.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            hog.computeBlockHistograms(cv::gpu::GpuMat(img));
+            compare_inner_parts(hog.block_hists, descriptors.rowRange(2, 3));
+
+            img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/negative2.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            hog.computeBlockHistograms(cv::gpu::GpuMat(img));
+            compare_inner_parts(hog.block_hists, descriptors.rowRange(3, 4));
+
+            img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/positive3.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            hog.computeBlockHistograms(cv::gpu::GpuMat(img));
+            compare_inner_parts(hog.block_hists, descriptors.rowRange(4, 5));
+
+            img_rgb = cv::imread(std::string(ts->get_data_path()) + "hog/negative3.png");
+            CHECK(!img_rgb.empty(), CvTS::FAIL_MISSING_TEST_DATA);
+            cv::cvtColor(img_rgb, img, CV_BGR2BGRA);
+            hog.computeBlockHistograms(cv::gpu::GpuMat(img));
+            compare_inner_parts(hog.block_hists, descriptors.rowRange(5, 6));
+        }
+        catch (const cv::Exception& e)
+        {
+            if (!check_and_treat_gpu_exception(e, ts)) throw;
+            return;
+        }
+    }
+
+    // Does not compare border value, as interpolation leads to delta
+    void compare_inner_parts(cv::Mat d1, cv::Mat d2)
+    {
+        for (int i = 1; i < blocks_per_win_y - 1; ++i)
+            for (int j = 1; j < blocks_per_win_x - 1; ++j)
+                for (int k = 0; k < block_hist_size; ++k)
+                {
+                    float a = d1.at<float>(0, (i * blocks_per_win_x + j) * block_hist_size);
+                    float b = d2.at<float>(0, (i * blocks_per_win_x + j) * block_hist_size);
+                    CHECK(a == b, CvTS::FAIL_INVALID_OUTPUT)
+                }
+    }
+
+    int wins_per_img_x;
+    int wins_per_img_y;
+    int blocks_per_win_x;
+    int blocks_per_win_y;
+    int block_hist_size;
+} gpu_hog_get_descriptors_test;
+