Merge branch 'mosift'

Josh Klontz
2 parents 2d868e20 f3357e59
Showing 5 changed files with 119 additions and 2 deletions
openbr/plugins/algorithms.cpp
openbr/plugins/hist.cpp
openbr/plugins/keypoint.cpp
openbr/plugins/misc.cpp
openbr/plugins/opticalflow.cpp
@@ -49,6 +49,10 @@ class AlgorithmsInitializer : public Initializer
         Globals->abbreviations.insert("DisplayVideo", "Stream([FPSLimit(30)+Show(false,[FrameNumber])+Discard])");
         Globals->abbreviations.insert("PerFrameDetection", "Stream([SaveMat(original)+Cvt(Gray)+Cascade(FrontalFace)+ASEFEyes+RestoreMat(original)+Draw(inPlace=true),Show(false,[FrameNumber])+Discard])");
         Globals->abbreviations.insert("AgeGenderDemo", "Stream([SaveMat(original)+Cvt(Gray)+Cascade(FrontalFace)+Expand+<FaceClassificationRegistration>+<FaceClassificationExtraction>+(<AgeRegressor>+Rename(Subject,Age)+Discard)/(<GenderClassifier>+Rename(Subject,Gender)+Discard)+RestoreMat(original)+Draw(inPlace=true)+DrawPropertiesPoint([Age,Gender],Affine_0,inPlace=true)+SaveMat(original)+Discard+Contract,RestoreMat(original)+FPSCalc+Show(false,[AvgFPS,Age,Gender])+Discard])");
+        Globals->abbreviations.insert("BoVW", "Flatten+CatRows+KMeans(500)+Hist(500)");
+        Globals->abbreviations.insert("HOF", "Stream([KeyPointDetector(SIFT),AggregateFrames(2)+OpticalFlow,ROI,HoGDescriptor])+BoVW");
+        Globals->abbreviations.insert("HoG", "Stream([KeyPointDetector(SIFT),ROI,HoGDescriptor])+BoVW");
+        Globals->abbreviations.insert("HoGHOF", "Stream([KeyPointDetector(SIFT),AggregateFrames(2),(OpticalFlow+ROI+HoGDescriptor)/(First+ROI+HoGDescriptor),Cat])+BoVW");
  
         // Generic Image Processing
         Globals->abbreviations.insert("SIFT", "Open+KeyPointDetector(SIFT)+KeyPointDescriptor(SIFT):KeyPointMatcher(BruteForce)");
@@ -52,8 +52,12 @@ class HistTransform : public UntrainableTransform
             int histSize[] = {dims};
             float range[] = {min, max};
             const float* ranges[] = {range};
-            Mat hist;
-            calcHist(&mv[i], 1, channels, Mat(), hist, 1, histSize, ranges);
+            Mat hist, chan = mv[i];
+            // calcHist requires F or U, might as well convert just in case
+            if (mv[i].depth() != CV_8U || mv[i].depth() == CV_32F)
+                mv[i].convertTo(chan, CV_32FC1);
+//            qDebug("clust r %i c %i ch %i", chan.rows, chan.cols, chan.channels());
+            calcHist(&chan, 1, channels, Mat(), hist, 1, histSize, ranges);
             memcpy(m.ptr(i), hist.ptr(), dims * sizeof(float));
         }
  
@@ -15,6 +15,7 @@
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
  
 #include <opencv2/features2d/features2d.hpp>
+#include <opencv2/objdetect/objdetect.hpp>
 #include <opencv2/nonfree/nonfree.hpp>
 #include "openbr_internal.h"
 #include "openbr/core/opencvutils.h"
@@ -178,6 +179,33 @@ BR_REGISTER(Transform, SIFTDescriptorTransform)
  
 /*!
  * \ingroup transforms
+ * \brief OpenCV HOGDescriptor wrapper
+ * \author Austin Blanton \cite imaus10
+ */
+class HoGDescriptorTransform : public UntrainableTransform
+{
+    Q_OBJECT
+
+    HOGDescriptor hog;
+
+    void project(const Template &src, Template &dst) const
+    {
+        std::vector<float> descriptorVals;
+        std::vector<Point> locations;
+        Size winStride = Size(0,0);
+        Size padding = Size(0,0);
+        foreach (const Mat &rect, src) {
+            hog.compute(rect, descriptorVals, winStride, padding, locations);
+            Mat HoGFeats(descriptorVals, true);
+            dst += HoGFeats;
+        }
+    }
+};
+
+BR_REGISTER(Transform, HoGDescriptorTransform)
+
+/*!
+ * \ingroup transforms
  * \brief Add landmarks to the template in a grid layout
  * \author Josh Klontz \cite jklontz
  */
@@ -538,6 +538,32 @@ class EventTransform : public UntrainableMetaTransform
 };
 BR_REGISTER(Transform, EventTransform)
  
+/*!
+ * \ingroup transforms
+ * \brief Flattens a list of Templates into a single Template
+ * \author Austin Blanton \cite imaus10
+ */
+class FlattenTransform : public UntrainableMetaTransform
+{
+    Q_OBJECT
+
+    void project(const TemplateList &src, TemplateList &dst) const
+    {
+        Template flat;
+        foreach (const Template &tmpl, src)
+            flat.append(tmpl);
+        dst.append(flat);
+    }
+
+    void project(const Template &src, Template &dst) const
+    {
+        (void) src;
+        (void) dst;
+        qFatal("You shouldn't do that!");
+    }
+};
+BR_REGISTER(Transform, FlattenTransform)
+
 }
  
 #include "misc.moc"
+#include <opencv2/video/tracking.hpp>
+#include "openbr_internal.h"
+
+using namespace cv;
+
+namespace br
+{
+
+/*!
+ * \ingroup transforms
+ * \brief Gets a one-channel dense optical flow from two images
+ * \author Austin Blanton \cite imaus10
+ */
+class OpticalFlowTransform : public UntrainableTransform
+{
+    Q_OBJECT
+    Q_PROPERTY(double pyr_scale READ get_pyr_scale WRITE set_pyr_scale RESET reset_pyr_scale STORED false)
+    Q_PROPERTY(int levels READ get_levels WRITE set_levels RESET reset_levels STORED false)
+    Q_PROPERTY(int winsize READ get_winsize WRITE set_winsize RESET reset_winsize STORED false)
+    Q_PROPERTY(int iterations READ get_iterations WRITE set_iterations RESET reset_iterations STORED false)
+    Q_PROPERTY(int poly_n READ get_poly_n WRITE set_poly_n RESET reset_poly_n STORED false)
+    Q_PROPERTY(double poly_sigma READ get_poly_sigma WRITE set_poly_sigma RESET reset_poly_sigma STORED false)
+    Q_PROPERTY(int flags READ get_flags WRITE set_flags RESET reset_flags STORED false)
+    // these defaults are optimized for KTH
+    BR_PROPERTY(double, pyr_scale, 0.1)
+    BR_PROPERTY(int, levels, 1)
+    BR_PROPERTY(int, winsize, 5)
+    BR_PROPERTY(int, iterations, 10)
+    BR_PROPERTY(int, poly_n, 7)
+    BR_PROPERTY(double, poly_sigma, 1.1)
+    BR_PROPERTY(int, flags, 0)
+
+    void project(const Template &src, Template &dst) const
+    {
+        // get the two images put there by AggregateFrames
+        Mat prevImg = src[0];
+        Mat nextImg = src[1];
+        Mat flow;
+        calcOpticalFlowFarneback(prevImg, nextImg, flow, pyr_scale, levels, winsize, iterations, poly_n, poly_sigma, flags);
+        
+        // the result is two channels
+        std::vector<Mat> channels(2);
+        split(flow, channels);
+        Mat flowOneCh;
+        magnitude(channels[0], channels[1], flowOneCh);
+
+        dst += flowOneCh;
+    }
+};
+
+BR_REGISTER(Transform, OpticalFlowTransform)
+
+} // namespace br
+
+#include "opticalflow.moc"