minor optimization of SURF_GPU (reduce memory transfers, use structure of arrays…

minor optimization of SURF_GPU (reduce memory transfers, use structure of arrays instead of array of structures)

modules/gpu/doc/feature_detection_and_description.rst

@@ -15,6 +15,17 @@ This class is used for extracting Speeded Up Robust Features (SURF) from an imag
     class SURF_GPU : public CvSURFParams
     {
     public:
+        enum KeypointLayout 
+        {
+            SF_X = 0,
+            SF_Y,
+            SF_LAPLACIAN,
+            SF_SIZE,
+            SF_DIR,
+            SF_HESSIAN,
+            SF_FEATURE_STRIDE
+        };
+
         //! the default constructor
         SURF_GPU();
         //! the full constructor taking all the necessary parameters
@@ -67,22 +78,15 @@ This class is used for extracting Speeded Up Robust Features (SURF) from an imag
         GpuMat det, trace;
 
         GpuMat maxPosBuffer;
-        GpuMat featuresBuffer;
-        GpuMat keypointsBuffer;
     };
 
 
 The class ``SURF_GPU`` implements Speeded Up Robust Features descriptor. There is a fast multi-scale Hessian keypoint detector that can be used to find the keypoints (which is the default option). But the descriptors can also be computed for the user-specified keypoints. Only 8 bit grayscale images are supported.
 
-The class ``SURF_GPU`` can store results in the GPU and CPU memory. It provides functions to convert results between CPU and GPU version ( ``uploadKeypoints``, ``downloadKeypoints``, ``downloadDescriptors`` ). The format of CPU results is the same as ``SURF`` results. GPU results are stored in  ``GpuMat`` . The ``keypoints`` matrix is a one-row matrix of the ``CV_32FC6`` type. It contains 6 float values per feature: ``x, y, laplacian, size, dir, hessian`` .  The ``descriptors`` matrix is
-:math:`\texttt{nFeatures} \times \texttt{descriptorSize}` matrix with the ``CV_32FC1`` type.
+The class ``SURF_GPU`` can store results in the GPU and CPU memory. It provides functions to convert results between CPU and GPU version ( ``uploadKeypoints``, ``downloadKeypoints``, ``downloadDescriptors`` ). The format of CPU results is the same as ``SURF`` results. GPU results are stored in  ``GpuMat`` . The ``keypoints`` matrix is :math:`\texttt{nFeatures} \times 6` matrix with the ``CV_32FC1`` type. keypoints.ptr<float>(SF_X)[i] will contain x coordinate of i'th feature. keypoints.ptr<float>(SF_Y)[i] will contain y coordinate of i'th feature. keypoints.ptr<float>(SF_LAPLACIAN)[i] will contain laplacian sign of i'th feature. keypoints.ptr<float>(SF_SIZE)[i] will contain size of i'th feature. keypoints.ptr<float>(SF_DIR)[i] will contain orientation of i'th feature. keypoints.ptr<float>(SF_HESSIAN)[i] will contain response of i'th feature. The ``descriptors`` matrix is :math:`\texttt{nFeatures} \times \texttt{descriptorSize}` matrix with the ``CV_32FC1`` type.
 
 The class ``SURF_GPU`` uses some buffers and provides access to it. All buffers can be safely released between function calls.
 
-**Note:**
-
-By default for user provided keypoints the class ``SURF_GPU`` recalculates keypoint's orientation and returns reodered/filtered keypoints array and coresponding decriptors array.
-
 See Also: :c:type:`SURF`
 
 .. index:: gpu::BruteForceMatcher_GPU

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

@@ -1566,6 +1566,17 @@ namespace cv
         class CV_EXPORTS SURF_GPU : public CvSURFParams
         {
         public:
+            enum KeypointLayout 
+            {
+                SF_X = 0,
+                SF_Y,
+                SF_LAPLACIAN,
+                SF_SIZE,
+                SF_DIR,
+                SF_HESSIAN,
+                SF_FEATURE_STRIDE
+            };
+
             //! the default constructor
             SURF_GPU();
             //! the full constructor taking all the necessary parameters
@@ -1585,9 +1596,13 @@ namespace cv
             
             //! finds the keypoints using fast hessian detector used in SURF
             //! supports CV_8UC1 images
-            //! keypoints will have 1 row and type CV_32FC(6)
-            //! keypoints.at<float[6]>(1, i) contains i'th keypoint
-            //! format: (x, y, laplacian, size, dir, hessian)
+            //! keypoints will have nFeature cols and 6 rows
+            //! keypoints.ptr<float>(SF_X)[i] will contain x coordinate of i'th feature
+            //! keypoints.ptr<float>(SF_Y)[i] will contain y coordinate of i'th feature
+            //! keypoints.ptr<float>(SF_LAPLACIAN)[i] will contain laplacian sign of i'th feature
+            //! keypoints.ptr<float>(SF_SIZE)[i] will contain size of i'th feature
+            //! keypoints.ptr<float>(SF_DIR)[i] will contain orientation of i'th feature
+            //! keypoints.ptr<float>(SF_HESSIAN)[i] will contain response of i'th feature
             void operator()(const GpuMat& img, const GpuMat& mask, GpuMat& keypoints);
             //! finds the keypoints and computes their descriptors. 
             //! Optionally it can compute descriptors for the user-provided keypoints and recompute keypoints direction
@@ -1611,8 +1626,6 @@ namespace cv
             GpuMat det, trace;
 
             GpuMat maxPosBuffer;
-            GpuMat featuresBuffer;
-            GpuMat keypointsBuffer;
         };
 
     }

modules/gpu/src/cuda/internal_shared.hpp

@@ -105,28 +105,7 @@ namespace cv
             const textureReference* tex; 
             cudaSafeCall( cudaGetTextureReference(&tex, name) ); 
             cudaSafeCall( cudaUnbindTexture(tex) );
-        }    
-
-        struct KeyPoint_GPU
-        {
-            float x;
-            float y;
-            float laplacian;
-            float size;
-            float dir;
-            float hessian;
-        };
-
-        enum KeypointLayout 
-        {
-            SF_X,
-            SF_Y,
-            SF_LAPLACIAN,
-            SF_SIZE,
-            SF_DIR,
-            SF_HESSIAN,
-            SF_FEATURE_STRIDE
-        };
+        }
     }
 }