Merge branch 'sp_dev' into saliency_shengxin_gsoc2017

Author: Masterqsx

modules/saliency/include/opencv2/saliency/saliencySpecializedClasses.hpp

@@ -302,6 +302,53 @@ class CV_EXPORTS_W MotionSaliencyBinWangApr2014 : public MotionSaliency
 
 };
 
+/** @brief the Deep Gaze 1 Saliency approach from
+
+This method use the convolution layers of the pretrained AlexNet, linear combination, center bias and softmax to generate saliency map
+*/
+class CV_EXPORTS_W DiscriminantSaliency : public MotionSaliency
+{
+private:
+    Size imgProcessingSize;
+    unsigned hiddenSpaceDimension;
+    unsigned centerSize;
+    unsigned windowSize;
+    unsigned patchSize;
+    unsigned temporalSize;
+    unsigned stride;
+public:
+    struct DT
+    {
+        Mat A;
+        Mat C;
+        Mat Q;
+        Mat R;
+        Mat S;
+        Mat MU;
+        double VAR;
+    };
+//    DiscriminantSaliency();
+    DiscriminantSaliency(unsigned = 1, Size = Size(127, 127), unsigned = 10, unsigned = 8, unsigned = 96, unsigned = 400, unsigned = 11);
+    virtual ~DiscriminantSaliency();
+    CV_WRAP static Ptr<DeepGaze1> create()
+    {
+        return makePtr<DeepGaze1>();
+    }
+    CV_WRAP bool computeSaliency( InputArray image, OutputArray saliencyMap )
+    {
+        if( image.empty() )
+            return false;
+        return computeSaliencyImpl( image, saliencyMap );
+    }
+    void dynamicTextureEstimator( const Mat, DT& );
+    void patchGenerator( const std::vector<Mat>& img_sq, unsigned index, unsigned r, unsigned c, Mat& center, Mat& surround, Mat& all );
+    std::vector<Mat> saliencyMapGenerator( std::vector<Mat>, std::vector<Mat>& );
+    void saliencyMapVisualize( InputArray _saliencyMap );
+protected:
+    bool computeSaliencyImpl( InputArray image, OutputArray saliencyMap );
+    double KLdivDT( const DT&, const DT& );
+};
+
 /************************************ Specific Objectness Specialized Classes ************************************/
 
 /**

modules/saliency/samples/DiscriminantSaliencySample.cpp

@@ -0,0 +1,64 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+
+#include <opencv2/core.hpp>
+#include <opencv2/imgproc.hpp>
+#include <opencv2/highgui.hpp>
+#include <opencv2/saliency.hpp>
+#include <vector>
+#include <string>
+#include <iostream>
+#include <fstream>
+
+using namespace std;
+using namespace cv;
+using namespace saliency;
+
+
+int main(int argc, char* argv[])
+{
+    const char *keys =
+            "{ help h usage ? |     | show this message }"
+            "{ start_frame    |0    | start frame index }"
+            "{ length         |12   | # of frames video contain   }"
+            "{ default        |1    | use default deep net(AlexNet) and default weights }"
+            "{ video_name     |skiing| the name of video in UCSD background subtraction }"
+            "{ img_folder_path|JPEGS| path to folder with frames }"
+            "{ res_level      |  3  | resolution level of output saliency map. Suggested Range [0, 4]. The higher the level is, the fast the processing is, the lower the resolution is }";
+
+    CommandLineParser parser(argc, argv, keys);
+    if (parser.has("help"))
+    {
+        parser.printMessage();
+        return 0;
+    }
+    vector<Mat> img_sq;
+    DiscriminantSaliency t;
+    if ( parser.get<bool>( "default" ) )
+    {
+        t = DiscriminantSaliency();
+    }
+    else
+    {
+        t = DiscriminantSaliency(parser.get<int>( "res_level" ));
+    }
+    for ( unsigned i = 1; i < parser.get<unsigned>( "length" ); i++ )
+    {
+        char index[256] = {0};
+        sprintf(index, "%d", i + parser.get<int>( "start_frame" ));
+        Mat temp = imread(parser.get<string>("img_folder_path") + "/" + parser.get<string>("video_name") + "/frame_" + index + ".jpg", 0);
+        //Mat temp = imread(string("JPEGS/traffic/frame_") + index + ".jpg", 0);
+        //resize(temp, temp, Size(127, 127));
+        img_sq.push_back(temp);
+    }
+    vector<Mat> saliency_sq;
+    t.computeSaliency(img_sq, saliency_sq);
+    for ( unsigned i = 0; i < saliency_sq.size(); i++ )
+    {
+       resize(saliency_sq[i], saliency_sq[i], Size(1024, 768));
+       t.saliencyMapVisualize(saliency_sq[i]);
+    }
+    return 0;
+} //main