Extend to support not only multiple scales but also multiple angles

[hercules2020/kcf.git] / src / threadctx.hpp

diff --git a/src/threadctx.hpp b/src/threadctx.hpp
index 85529ddf7e193e57cb3c8d0a80b57d6df8707632..0b23707172cbbc1033e5de02980c076041f8834f 100644 (file)
--- a/src/threadctx.hpp
+++ b/src/threadctx.hpp
@@ -4,68 +4,97 @@
 #include <future>
 #include "dynmem.hpp"
 #include "kcf.h"
-
-#ifdef CUFFT
-#include "complexmat.cuh"
-#else
 #include "complexmat.hpp"
-#endif
+#include <vector>
 
 class KCF_Tracker;
 
+template <typename T>
+class ScaleRotVector : public std::vector<T> {
+public:
+    ScaleRotVector(const std::vector<double> &scales, const std::vector<double> &angles)
+        : scales(scales)
+        , angles(angles)
+    {}
+
+    uint getIdx(uint scale_idx, uint angle_idx) const { return angles.size() * scale_idx + angle_idx; }
+    uint getScaleIdx(uint idx) const { return idx / angles.size(); }
+    uint getAngleIdx(uint idx) const { return idx % angles.size(); }
+    T& operator()(uint scale_idx, uint angle_idx) { return std::vector<T>::at(getIdx(scale_idx, angle_idx)); }
+    double scale(uint idx) const { return scales[getScaleIdx(idx)]; }
+    double angle(uint idx) const { return angles[getAngleIdx(idx)]; }
+private:
+    const std::vector<double> scales, angles;
+};
+
 struct ThreadCtx {
   public:
-    ThreadCtx(cv::Size roi, uint num_channels, uint num_of_scales
-#ifndef BIG_BATCH
+    ThreadCtx(cv::Size roi, uint num_features
+#ifdef BIG_BATCH
+              , const std::vector<double> &scales
+              , const std::vector<double> &angles
+#else
               , double scale
+              , double angle
 #endif
              )
         : roi(roi)
-        , num_channels(num_channels)
-        , num_of_scales(num_of_scales)
-#ifndef BIG_BATCH
+        , num_features(num_features)
+        , num_scales(IF_BIG_BATCH(scales.size(), 1))
+        , num_angles(IF_BIG_BATCH(angles.size(), 1))
+#ifdef BIG_BATCH
+        , max(scales, angles)
+        , dbg_patch(scales, angles)
+        {
+            max.resize(scales.size() * angles.size());
+            dbg_patch.resize(scales.size() * angles.size());
+        }
+#else
         , scale(scale)
+        , angle(angle)
+        {}
 #endif
-    {}
+
 
     ThreadCtx(ThreadCtx &&) = default;
 
+    void track(const KCF_Tracker &kcf, cv::Mat &input_rgb, cv::Mat &input_gray);
 private:
     cv::Size roi;
-    uint num_channels;
-    uint num_of_scales;
+    uint num_features;
+    uint num_scales;
+    uint num_angles;
     cv::Size freq_size = Fft::freq_size(roi);
 
+    MatScaleFeats patch_feats{num_scales * num_angles, num_features, roi};
+    MatScaleFeats temp{num_scales * num_angles, num_features, roi};
+
+    KCF_Tracker::GaussianCorrelation gaussian_correlation{num_scales * num_angles, num_features, roi};
+
+    MatScales ifft2_res{num_scales * num_angles, roi};
+
+    ComplexMat zf{uint(freq_size.height), uint(freq_size.width), num_features, num_scales * num_angles};
+    ComplexMat kzf{uint(freq_size.height), uint(freq_size.width), 1, num_scales * num_angles};
+
 public:
 #ifdef ASYNC
     std::future<void> async_res;
 #endif
 
-    KCF_Tracker::GaussianCorrelation gaussian_correlation{roi, num_of_scales, num_channels};
-
-#if defined(CUFFT) || defined(FFTW) // TODO: Why this ifdef?
-    MatDynMem in_all{roi.height * int(num_of_scales), roi.width, CV_32F};
-#else
-    MatDynMem in_all{roi, CV_32F};
-#endif
-    MatDynMem fw_all{roi.height * int(num_channels), roi.width, CV_32F};
-    MatDynMem ifft2_res{roi, CV_32FC(int(num_channels))};
-    MatDynMem response{roi, CV_32FC(int(num_of_scales))};
-
-    ComplexMat zf{uint(freq_size.height), uint(freq_size.width), num_channels, num_of_scales};
-    ComplexMat kzf{uint(freq_size.height), uint(freq_size.width), num_of_scales};
+    MatScales response{num_scales * num_angles, roi};
 
-    // Variables used during non big batch mode and in big batch mode with ThreadCtx in p_threadctxs in kcf  on zero index.
-    cv::Point2i max_loc;
-    double max_val, max_response;
+    struct Max {
+        cv::Point2i loc;
+        double response;
+    };
 
 #ifdef BIG_BATCH
-    // Stores value of responses, location of maximal response and response maps for each scale
-    std::vector<double> max_responses = std::vector<double>(num_of_scales);
-    std::vector<cv::Point2i> max_locs = std::vector<cv::Point2i>(num_of_scales);
-    std::vector<cv::Mat> response_maps = std::vector<cv::Mat>(num_of_scales);
+    ScaleRotVector<Max> max;
+    ScaleRotVector<cv::Mat> dbg_patch; // images for visual debugging
 #else
-    const double scale;
+    Max max;
+    const double scale, angle;
+    cv::Mat dbg_patch; // image for visual debugging
 #endif
 };