Merge pull request #1195 from woodychow:multithread_sift_findScaleSpaceExtrema

modules/xfeatures2d/src/sift.cpp

@@ -569,31 +569,53 @@ static bool adjustLocalExtrema( const std::vector<Mat>& dog_pyr, KeyPoint& kpt,
 }
 
 
-//
+class findScaleSpaceExtremaComputer : public ParallelLoopBody
-// Detects features at extrema in DoG scale space.  Bad features are discarded
-// based on contrast and ratio of principal curvatures.
-void SIFT_Impl::findScaleSpaceExtrema( const std::vector<Mat>& gauss_pyr, const std::vector<Mat>& dog_pyr,
-                                  std::vector<KeyPoint>& keypoints ) const
 {
-    int nOctaves = (int)gauss_pyr.size()/(nOctaveLayers + 3);
+public:
-    int threshold = cvFloor(0.5 * contrastThreshold / nOctaveLayers * 255 * SIFT_FIXPT_SCALE);
+    findScaleSpaceExtremaComputer(
-    const int n = SIFT_ORI_HIST_BINS;
+        int _o,
-    float hist[n];
+        int _i,
-    KeyPoint kpt;
+        int _threshold,
+        int _idx,
+        int _step,
+        int _cols,
+        int _nOctaveLayers,
+        double _contrastThreshold,
+        double _edgeThreshold,
+        double _sigma,
+        const std::vector<Mat>& _gauss_pyr,
+        const std::vector<Mat>& _dog_pyr,
+        TLSData<std::vector<KeyPoint> > &_tls_kpts_struct)
+
+        : o(_o),
+          i(_i),
+          threshold(_threshold),
+          idx(_idx),
+          step(_step),
+          cols(_cols),
+          nOctaveLayers(_nOctaveLayers),
+          contrastThreshold(_contrastThreshold),
+          edgeThreshold(_edgeThreshold),
+          sigma(_sigma),
+          gauss_pyr(_gauss_pyr),
+          dog_pyr(_dog_pyr),
+          tls_kpts_struct(_tls_kpts_struct) { }
+    void operator()( const cv::Range& range ) const
+    {
+        const int begin = range.start;
+        const int end = range.end;
 
-    keypoints.clear();
+        static const int n = SIFT_ORI_HIST_BINS;
+        float hist[n];
 
-    for( int o = 0; o < nOctaves; o++ )
-        for( int i = 1; i <= nOctaveLayers; i++ )
-        {
-            int idx = o*(nOctaveLayers+2)+i;
         const Mat& img = dog_pyr[idx];
         const Mat& prev = dog_pyr[idx-1];
         const Mat& next = dog_pyr[idx+1];
-            int step = (int)img.step1();
-            int rows = img.rows, cols = img.cols;
 
-            for( int r = SIFT_IMG_BORDER; r < rows-SIFT_IMG_BORDER; r++)
+        std::vector<KeyPoint> *tls_kpts = tls_kpts_struct.get();
+
+        KeyPoint kpt;
+        for( int r = begin; r < end; r++)
         {
             const sift_wt* currptr = img.ptr<sift_wt>(r);
             const sift_wt* prevptr = prev.ptr<sift_wt>(r);
@@ -648,7 +670,8 @@ void SIFT_Impl::findScaleSpaceExtrema( const std::vector<Mat>& gauss_pyr, const
                             kpt.angle = 360.f - (float)((360.f/n) * bin);
                             if(std::abs(kpt.angle - 360.f) < FLT_EPSILON)
                                 kpt.angle = 0.f;
-                                keypoints.push_back(kpt);
+                            {
+                                tls_kpts->push_back(kpt);
                             }
                         }
                     }
@@ -656,6 +679,55 @@ void SIFT_Impl::findScaleSpaceExtrema( const std::vector<Mat>& gauss_pyr, const
             }
         }
     }
+private:
+    int o, i;
+    int threshold;
+    int idx, step, cols;
+    int nOctaveLayers;
+    double contrastThreshold;
+    double edgeThreshold;
+    double sigma;
+    const std::vector<Mat>& gauss_pyr;
+    const std::vector<Mat>& dog_pyr;
+    TLSData<std::vector<KeyPoint> > &tls_kpts_struct;
+};
+
+//
+// Detects features at extrema in DoG scale space.  Bad features are discarded
+// based on contrast and ratio of principal curvatures.
+void SIFT_Impl::findScaleSpaceExtrema( const std::vector<Mat>& gauss_pyr, const std::vector<Mat>& dog_pyr,
+                                  std::vector<KeyPoint>& keypoints ) const
+{
+    const int nOctaves = (int)gauss_pyr.size()/(nOctaveLayers + 3);
+    const int threshold = cvFloor(0.5 * contrastThreshold / nOctaveLayers * 255 * SIFT_FIXPT_SCALE);
+
+    keypoints.clear();
+    TLSData<std::vector<KeyPoint> > tls_kpts_struct;
+
+    for( int o = 0; o < nOctaves; o++ )
+        for( int i = 1; i <= nOctaveLayers; i++ )
+        {
+            const int idx = o*(nOctaveLayers+2)+i;
+            const Mat& img = dog_pyr[idx];
+            const int step = (int)img.step1();
+            const int rows = img.rows, cols = img.cols;
+
+            parallel_for_(Range(SIFT_IMG_BORDER, rows-SIFT_IMG_BORDER),
+                findScaleSpaceExtremaComputer(
+                    o, i, threshold, idx, step, cols,
+                    nOctaveLayers,
+                    contrastThreshold,
+                    edgeThreshold,
+                    sigma,
+                    gauss_pyr, dog_pyr, tls_kpts_struct));
+        }
+
+    std::vector<std::vector<KeyPoint>*> kpt_vecs;
+    tls_kpts_struct.gather(kpt_vecs);
+    for (size_t i = 0; i < kpt_vecs.size(); ++i) {
+        keypoints.insert(keypoints.end(), kpt_vecs[i]->begin(), kpt_vecs[i]->end());
+    }
+}
 
 
 static void calcSIFTDescriptor( const Mat& img, Point2f ptf, float ori, float scl,
@@ -1081,7 +1153,7 @@ void SIFT_Impl::detectAndCompute(InputArray _image, InputArray _mask,
     {
         //t = (double)getTickCount();
         findScaleSpaceExtrema(gpyr, dogpyr, keypoints);
-        KeyPointsFilter::removeDuplicated( keypoints );
+        KeyPointsFilter::removeDuplicatedSorted( keypoints );
 
         if( nfeatures > 0 )
             KeyPointsFilter::retainBest(keypoints, nfeatures);