#include "kcf.h"
#include <numeric>
#include <thread>
-#include <future>
#include <algorithm>
+#include "threadctx.hpp"
+#include "debug.h"
#ifdef FFTW
- #include "fft_fftw.h"
- #define FFT Fftw
+#include "fft_fftw.h"
+#define FFT Fftw
#elif defined(CUFFT)
- #include "fft_cufft.h"
- #define FFT cuFFT
+#include "fft_cufft.h"
+#define FFT cuFFT
#else
- #include "fft_opencv.h"
- #define FFT FftOpencv
+#include "fft_opencv.h"
+#define FFT FftOpencv
#endif
#ifdef OPENMP
#include <omp.h>
-#endif //OPENMP
+#endif // OPENMP
-#define DEBUG_PRINT(obj) if (m_debug) {std::cout << #obj << " @" << __LINE__ << std::endl << (obj) << std::endl;}
-#define DEBUG_PRINTM(obj) if (m_debug) {std::cout << #obj << " @" << __LINE__ << " " << (obj).size() << " CH: " << (obj).channels() << std::endl << (obj) << std::endl;}
+DbgTracer __dbgTracer;
-KCF_Tracker::KCF_Tracker(double padding, double kernel_sigma, double lambda, double interp_factor, double output_sigma_factor, int cell_size) :
- fft(*new FFT()),
- p_padding(padding), p_output_sigma_factor(output_sigma_factor), p_kernel_sigma(kernel_sigma),
- p_lambda(lambda), p_interp_factor(interp_factor), p_cell_size(cell_size) {}
+template <typename T>
+T clamp(const T& n, const T& lower, const T& upper)
+{
+ return std::max(lower, std::min(n, upper));
+}
+
+template <typename T>
+void clamp2(T& n, const T& lower, const T& upper)
+{
+ n = std::max(lower, std::min(n, upper));
+}
+
+#if CV_MAJOR_VERSION < 3
+template<typename _Tp> static inline
+cv::Size_<_Tp> operator / (const cv::Size_<_Tp>& a, _Tp b)
+{
+ return cv::Size_<_Tp>(a.width / b, a.height / b);
+}
+#endif
-KCF_Tracker::KCF_Tracker()
- : fft(*new FFT()) {}
+class Kcf_Tracker_Private {
+ friend KCF_Tracker;
+ std::vector<ThreadCtx> threadctxs;
+};
+
+KCF_Tracker::KCF_Tracker(double padding, double kernel_sigma, double lambda, double interp_factor,
+ double output_sigma_factor, int cell_size)
+ : p_cell_size(cell_size), fft(*new FFT()), p_padding(padding), p_output_sigma_factor(output_sigma_factor), p_kernel_sigma(kernel_sigma),
+ p_lambda(lambda), p_interp_factor(interp_factor), d(*new Kcf_Tracker_Private)
+{
+}
+
+KCF_Tracker::KCF_Tracker() : fft(*new FFT()), d(*new Kcf_Tracker_Private) {}
KCF_Tracker::~KCF_Tracker()
{
delete &fft;
-#ifdef CUFFT
- for (int i = 0;i < p_num_scales;++i) {
- CudaSafeCall(cudaFreeHost(p_scale_vars[i].xf_sqr_norm));
- CudaSafeCall(cudaFreeHost(p_scale_vars[i].yf_sqr_norm));
- CudaSafeCall(cudaFreeHost(p_scale_vars[i].data_i_1ch));
- CudaSafeCall(cudaFreeHost(p_scale_vars[i].data_i_features));
- CudaSafeCall(cudaFree(p_scale_vars[i].gauss_corr_res_d));
- CudaSafeCall(cudaFreeHost(p_scale_vars[i].rot_labels_data));
- CudaSafeCall(cudaFreeHost(p_scale_vars[i].data_features));
- }
-#else
- for (int i = 0;i < p_num_scales;++i) {
- free(p_scale_vars[i].xf_sqr_norm);
- free(p_scale_vars[i].yf_sqr_norm);
+ delete &d;
+}
+
+void KCF_Tracker::train(cv::Mat input_rgb, cv::Mat input_gray, double interp_factor)
+{
+ TRACE("");
+
+ // obtain a sub-window for training
+ // TODO: Move Mats outside from here
+ MatScaleFeats patch_feats(1, p_num_of_feats, feature_size);
+ DEBUG_PRINT(patch_feats);
+ MatScaleFeats temp(1, p_num_of_feats, feature_size);
+ get_features(input_rgb, input_gray, p_current_center.x, p_current_center.y,
+ p_windows_size.width, p_windows_size.height,
+ p_current_scale).copyTo(patch_feats.scale(0));
+ DEBUG_PRINT(patch_feats);
+ fft.forward_window(patch_feats, p_xf, temp);
+ DEBUG_PRINTM(p_xf);
+ p_model_xf = p_model_xf * (1. - interp_factor) + p_xf * interp_factor;
+ DEBUG_PRINTM(p_model_xf);
+
+ ComplexMat alphaf_num, alphaf_den;
+
+ if (m_use_linearkernel) {
+ ComplexMat xfconj = p_xf.conj();
+ alphaf_num = xfconj.mul(p_yf);
+ alphaf_den = (p_xf * xfconj);
+ } else {
+ // Kernel Ridge Regression, calculate alphas (in Fourier domain)
+ cv::Size sz(Fft::freq_size(feature_size));
+ ComplexMat kf(sz.height, sz.width, 1);
+ (*gaussian_correlation)(kf, p_model_xf, p_model_xf, p_kernel_sigma, true, *this);
+ DEBUG_PRINTM(kf);
+ p_model_alphaf_num = p_yf * kf;
+ p_model_alphaf_den = kf * (kf + p_lambda);
}
-#endif
+ p_model_alphaf = p_model_alphaf_num / p_model_alphaf_den;
+ DEBUG_PRINTM(p_model_alphaf);
+ // p_model_alphaf = p_yf / (kf + p_lambda); //equation for fast training
}
-void KCF_Tracker::init(cv::Mat &img, const cv::Rect & bbox, int fit_size_x, int fit_size_y)
+static int round_pw2_down(int x)
+{
+ for (int i = 1; i < 32; i <<= 1)
+ x |= x >> i;
+ x++;
+ return x >> 1;
+}
+
+
+void KCF_Tracker::init(cv::Mat &img, const cv::Rect &bbox, int fit_size_x, int fit_size_y)
{
- //check boundary, enforce min size
+ __dbgTracer.debug = m_debug;
+ TRACE("");
+
+ // check boundary, enforce min size
double x1 = bbox.x, x2 = bbox.x + bbox.width, y1 = bbox.y, y2 = bbox.y + bbox.height;
if (x1 < 0) x1 = 0.;
- if (x2 > img.cols-1) x2 = img.cols - 1;
+ if (x2 > img.cols - 1) x2 = img.cols - 1;
if (y1 < 0) y1 = 0;
- if (y2 > img.rows-1) y2 = img.rows - 1;
+ if (y2 > img.rows - 1) y2 = img.rows - 1;
- if (x2-x1 < 2*p_cell_size) {
- double diff = (2*p_cell_size -x2+x1)/2.;
- if (x1 - diff >= 0 && x2 + diff < img.cols){
+ if (x2 - x1 < 2 * p_cell_size) {
+ double diff = (2 * p_cell_size - x2 + x1) / 2.;
+ if (x1 - diff >= 0 && x2 + diff < img.cols) {
x1 -= diff;
x2 += diff;
- } else if (x1 - 2*diff >= 0) {
- x1 -= 2*diff;
+ } else if (x1 - 2 * diff >= 0) {
+ x1 -= 2 * diff;
} else {
- x2 += 2*diff;
+ x2 += 2 * diff;
}
}
- if (y2-y1 < 2*p_cell_size) {
- double diff = (2*p_cell_size -y2+y1)/2.;
- if (y1 - diff >= 0 && y2 + diff < img.rows){
+ if (y2 - y1 < 2 * p_cell_size) {
+ double diff = (2 * p_cell_size - y2 + y1) / 2.;
+ if (y1 - diff >= 0 && y2 + diff < img.rows) {
y1 -= diff;
y2 += diff;
- } else if (y1 - 2*diff >= 0) {
- y1 -= 2*diff;
+ } else if (y1 - 2 * diff >= 0) {
+ y1 -= 2 * diff;
} else {
- y2 += 2*diff;
+ y2 += 2 * diff;
}
}
- p_pose.w = x2-x1;
- p_pose.h = y2-y1;
- p_pose.cx = x1 + p_pose.w/2.;
- p_pose.cy = y1 + p_pose.h /2.;
-
+ p_init_pose.w = x2 - x1;
+ p_init_pose.h = y2 - y1;
+ p_init_pose.cx = x1 + p_init_pose.w / 2.;
+ p_init_pose.cy = y1 + p_init_pose.h / 2.;
cv::Mat input_gray, input_rgb = img.clone();
- if (img.channels() == 3){
+ if (img.channels() == 3) {
cv::cvtColor(img, input_gray, CV_BGR2GRAY);
input_gray.convertTo(input_gray, CV_32FC1);
- }else
+ } else
img.convertTo(input_gray, CV_32FC1);
// don't need too large image
- if (p_pose.w * p_pose.h > 100.*100. && (fit_size_x == -1 || fit_size_y == -1)) {
- std::cout << "resizing image by factor of " << 1/p_downscale_factor << std::endl;
+ if (p_init_pose.w * p_init_pose.h > 100. * 100.) {
+ std::cout << "resizing image by factor of " << 1 / p_downscale_factor << std::endl;
p_resize_image = true;
- p_pose.scale(p_downscale_factor);
- cv::resize(input_gray, input_gray, cv::Size(0,0), p_downscale_factor, p_downscale_factor, cv::INTER_AREA);
- cv::resize(input_rgb, input_rgb, cv::Size(0,0), p_downscale_factor, p_downscale_factor, cv::INTER_AREA);
- } else if (!(fit_size_x == -1 && fit_size_y == -1)) {
- if (fit_size_x%p_cell_size != 0 || fit_size_y%p_cell_size != 0) {
- std::cerr << "Fit size does not fit to hog cell size. The dimensions have to be divisible by HOG cell size, which is: " << p_cell_size << std::endl;;
- std::exit(EXIT_FAILURE);
- }
- double tmp;
- if (( tmp = (p_pose.w * (1. + p_padding) / p_cell_size) * p_cell_size ) != fit_size_x)
- p_scale_factor_x = fit_size_x/tmp;
- if (( tmp = (p_pose.h * (1. + p_padding) / p_cell_size) * p_cell_size ) != fit_size_y)
- p_scale_factor_y = fit_size_y/tmp;
- std::cout << "resizing image horizontaly by factor of " << p_scale_factor_x
- << " and verticaly by factor of " << p_scale_factor_y << std::endl;
- p_fit_to_pw2 = true;
- p_pose.scale_x(p_scale_factor_x);
- p_pose.scale_y(p_scale_factor_y);
- if (p_scale_factor_x != 1 && p_scale_factor_y != 1) {
- if (p_scale_factor_x < 1 && p_scale_factor_y < 1) {
- cv::resize(input_gray, input_gray, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_AREA);
- cv::resize(input_rgb, input_rgb, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_AREA);
- } else {
- cv::resize(input_gray, input_gray, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_LINEAR);
- cv::resize(input_rgb, input_rgb, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_LINEAR);
- }
- }
+ p_init_pose.scale(p_downscale_factor);
+ cv::resize(input_gray, input_gray, cv::Size(0, 0), p_downscale_factor, p_downscale_factor, cv::INTER_AREA);
+ cv::resize(input_rgb, input_rgb, cv::Size(0, 0), p_downscale_factor, p_downscale_factor, cv::INTER_AREA);
}
- //compute win size + fit to fhog cell size
- p_windows_size[0] = round(p_pose.w * (1. + p_padding) / p_cell_size) * p_cell_size;
- p_windows_size[1] = round(p_pose.h * (1. + p_padding) / p_cell_size) * p_cell_size;
+ // compute win size + fit to fhog cell size
+ p_windows_size.width = round(p_init_pose.w * (1. + p_padding) / p_cell_size) * p_cell_size;
+ p_windows_size.height = round(p_init_pose.h * (1. + p_padding) / p_cell_size) * p_cell_size;
+
+ if (fit_size_x == 0 || fit_size_y == 0) {
+ // Round down to the next highest power of 2
+ fit_size = cv::Size(round_pw2_down(p_windows_size.width),
+ round_pw2_down(p_windows_size.height));
+ } else if (fit_size_x == -1 || fit_size_y == -1) {
+ fit_size = p_windows_size;
+ } else {
+ fit_size = cv::Size(fit_size_x, fit_size_y);
+ }
+
+ feature_size = fit_size / p_cell_size;
p_scales.clear();
- if (m_use_scale)
- for (int i = -p_num_scales/2; i <= p_num_scales/2; ++i)
- p_scales.push_back(std::pow(p_scale_step, i));
- else
- p_scales.push_back(1.);
+ for (int i = -int(p_num_scales) / 2; i <= int(p_num_scales) / 2; ++i)
+ p_scales.push_back(std::pow(p_scale_step, i));
#ifdef CUFFT
- if (p_windows_size[1]/p_cell_size*(p_windows_size[0]/p_cell_size/2+1) > 1024) {
+ if (Fft::freq_size(feature_size).area() > 1024) {
std::cerr << "Window after forward FFT is too big for CUDA kernels. Plese use -f to set "
- "the window dimensions so its size is less or equal to " << 1024*p_cell_size*p_cell_size*2+1 <<
- " pixels . Currently the size of the window is: " << p_windows_size[0] << "x" << p_windows_size[1] <<
- " which is " << p_windows_size[0]*p_windows_size[1] << " pixels. " << std::endl;
+ "the window dimensions so its size is less or equal to "
+ << 1024 * p_cell_size * p_cell_size * 2 + 1
+ << " pixels. Currently the size of the window is: " << fit_size
+ << " which is " << fit_size.area() << " pixels. " << std::endl;
std::exit(EXIT_FAILURE);
}
- if (m_use_linearkernel){
+ if (m_use_linearkernel) {
std::cerr << "cuFFT supports only Gaussian kernel." << std::endl;
std::exit(EXIT_FAILURE);
}
#endif
- p_num_of_feats = 31;
- if(m_use_color) p_num_of_feats += 3;
- if(m_use_cnfeat) p_num_of_feats += 10;
- p_roi_width = p_windows_size[0]/p_cell_size;
- p_roi_height = p_windows_size[1]/p_cell_size;
+#if defined(CUFFT) || defined(FFTW)
+ uint width = feature_size.width / 2 + 1;
+#else
+ uint width = feature_size.width;
+#endif
+ p_model_xf.create(feature_size.height, width, p_num_of_feats);
+ p_yf.create(feature_size.height, width, 1);
+ p_xf.create(feature_size.height, width, p_num_of_feats);
- for (int i = 0;i<p_num_scales;++i) {
- if (i == 0)
- p_scale_vars.push_back(Scale_vars(p_windows_size, p_cell_size, p_num_of_feats, &p_model_xf, &p_yf, true));
- else
- p_scale_vars.push_back(Scale_vars(p_windows_size, p_cell_size, p_num_of_feats));
- }
+#ifndef BIG_BATCH
+ for (auto scale: p_scales)
+ d.threadctxs.emplace_back(feature_size, p_num_of_feats, scale);
+#else
+ d.threadctxs.emplace_back(feature_size, p_num_of_feats, p_num_scales);
+#endif
+ gaussian_correlation.reset(new GaussianCorrelation(1, feature_size));
+
+ p_current_center = p_init_pose.center();
p_current_scale = 1.;
- double min_size_ratio = std::max(5.*p_cell_size/p_windows_size[0], 5.*p_cell_size/p_windows_size[1]);
- double max_size_ratio = std::min(floor((img.cols + p_windows_size[0]/3)/p_cell_size)*p_cell_size/p_windows_size[0], floor((img.rows + p_windows_size[1]/3)/p_cell_size)*p_cell_size/p_windows_size[1]);
+ double min_size_ratio = std::max(5. * p_cell_size / p_windows_size.width, 5. * p_cell_size / p_windows_size.height);
+ double max_size_ratio =
+ std::min(floor((img.cols + p_windows_size.width / 3) / p_cell_size) * p_cell_size / p_windows_size.width,
+ floor((img.rows + p_windows_size.height / 3) / p_cell_size) * p_cell_size / p_windows_size.height);
p_min_max_scale[0] = std::pow(p_scale_step, std::ceil(std::log(min_size_ratio) / log(p_scale_step)));
p_min_max_scale[1] = std::pow(p_scale_step, std::floor(std::log(max_size_ratio) / log(p_scale_step)));
- std::cout << "init: img size " << img.cols << " " << img.rows << std::endl;
- std::cout << "init: win size. " << p_windows_size[0] << " " << p_windows_size[1] << std::endl;
+ std::cout << "init: img size " << img.size() << std::endl;
+ std::cout << "init: win size " << p_windows_size;
+ if (p_windows_size != fit_size)
+ std::cout << " resized to " << fit_size;
+ std::cout << std::endl;
+ std::cout << "init: FFT size " << feature_size << std::endl;
std::cout << "init: min max scales factors: " << p_min_max_scale[0] << " " << p_min_max_scale[1] << std::endl;
- p_output_sigma = std::sqrt(p_pose.w*p_pose.h) * p_output_sigma_factor / static_cast<double>(p_cell_size);
+ p_output_sigma = std::sqrt(p_init_pose.w * p_init_pose.h * double(fit_size.area()) / p_windows_size.area())
+ * p_output_sigma_factor / p_cell_size;
- fft.init(p_windows_size[0]/p_cell_size, p_windows_size[1]/p_cell_size, p_num_of_feats, p_num_scales, m_use_big_batch);
- fft.set_window(cosine_window_function(p_windows_size[0]/p_cell_size, p_windows_size[1]/p_cell_size));
+ fft.init(feature_size.width, feature_size.height, p_num_of_feats, p_num_scales);
+ fft.set_window(MatDynMem(cosine_window_function(feature_size.width, feature_size.height)));
- //window weights, i.e. labels
- fft.forward(gaussian_shaped_labels(p_output_sigma, p_windows_size[0]/p_cell_size, p_windows_size[1]/p_cell_size), p_yf,
- m_use_cuda ? p_scale_vars[0].rot_labels_data_d: nullptr);
+ // window weights, i.e. labels
+ MatScales gsl(1, feature_size);
+ gaussian_shaped_labels(p_output_sigma, feature_size.width, feature_size.height).copyTo(gsl.plane(0));
+ fft.forward(gsl, p_yf);
DEBUG_PRINTM(p_yf);
- //obtain a sub-window for training initial model
- get_features(input_rgb, input_gray, p_pose.cx, p_pose.cy, p_windows_size[0], p_windows_size[1], p_scale_vars[0]);
- fft.forward_window(p_scale_vars[0].patch_feats, p_model_xf, p_scale_vars[0].fw_all, m_use_cuda ? p_scale_vars[0].data_features_d : nullptr);
- DEBUG_PRINTM(p_model_xf);
-
-
- if (m_use_linearkernel) {
- ComplexMat xfconj = p_model_xf.conj();
- p_model_alphaf_num = xfconj.mul(p_yf);
- p_model_alphaf_den = (p_model_xf * xfconj);
- } else {
- //Kernel Ridge Regression, calculate alphas (in Fourier domain)
- gaussian_correlation(p_scale_vars[0], p_model_xf, p_model_xf, p_kernel_sigma, true);
- DEBUG_PRINTM(p_scale_vars[0].kf);
- p_model_alphaf_num = p_yf * p_scale_vars[0].kf;
- DEBUG_PRINTM(p_model_alphaf_num);
- p_model_alphaf_den = p_scale_vars[0].kf * (p_scale_vars[0].kf + p_lambda);
- DEBUG_PRINTM(p_model_alphaf_den);
- }
- p_model_alphaf = p_model_alphaf_num / p_model_alphaf_den;
- DEBUG_PRINTM(p_model_alphaf);
-// p_model_alphaf = p_yf / (kf + p_lambda); //equation for fast training
+ // train initial model
+ train(input_rgb, input_gray, 1.0);
}
-void KCF_Tracker::setTrackerPose(BBox_c &bbox, cv::Mat & img, int fit_size_x, int fit_size_y)
+void KCF_Tracker::setTrackerPose(BBox_c &bbox, cv::Mat &img, int fit_size_x, int fit_size_y)
{
init(img, bbox.get_rect(), fit_size_x, fit_size_y);
}
void KCF_Tracker::updateTrackerPosition(BBox_c &bbox)
{
+ BBox_c tmp = bbox;
if (p_resize_image) {
- BBox_c tmp = bbox;
tmp.scale(p_downscale_factor);
- p_pose.cx = tmp.cx;
- p_pose.cy = tmp.cy;
- } else if (p_fit_to_pw2) {
- BBox_c tmp = bbox;
- tmp.scale_x(p_scale_factor_x);
- tmp.scale_y(p_scale_factor_y);
- p_pose.cx = tmp.cx;
- p_pose.cy = tmp.cy;
- } else {
- p_pose.cx = bbox.cx;
- p_pose.cy = bbox.cy;
}
+ p_current_center = tmp.center();
}
BBox_c KCF_Tracker::getBBox()
{
- BBox_c tmp = p_pose;
- tmp.w *= p_current_scale;
- tmp.h *= p_current_scale;
+ BBox_c tmp;
+ tmp.cx = p_current_center.x;
+ tmp.cy = p_current_center.y;
+ tmp.w = p_init_pose.w * p_current_scale;
+ tmp.h = p_init_pose.h * p_current_scale;
if (p_resize_image)
- tmp.scale(1/p_downscale_factor);
- if (p_fit_to_pw2) {
- tmp.scale_x(1/p_scale_factor_x);
- tmp.scale_y(1/p_scale_factor_y);
- }
+ tmp.scale(1 / p_downscale_factor);
return tmp;
}
-void KCF_Tracker::track(cv::Mat &img)
+double KCF_Tracker::getFilterResponse() const
{
- if (m_debug)
- std::cout << "NEW FRAME" << '\n';
- cv::Mat input_gray, input_rgb = img.clone();
- if (img.channels() == 3){
- cv::cvtColor(img, input_gray, CV_BGR2GRAY);
- input_gray.convertTo(input_gray, CV_32FC1);
- }else
- img.convertTo(input_gray, CV_32FC1);
+ return this->max_response;
+}
- // don't need too large image
+void KCF_Tracker::resizeImgs(cv::Mat &input_rgb, cv::Mat &input_gray)
+{
if (p_resize_image) {
cv::resize(input_gray, input_gray, cv::Size(0, 0), p_downscale_factor, p_downscale_factor, cv::INTER_AREA);
cv::resize(input_rgb, input_rgb, cv::Size(0, 0), p_downscale_factor, p_downscale_factor, cv::INTER_AREA);
- } else if (p_fit_to_pw2 && p_scale_factor_x != 1 && p_scale_factor_y != 1) {
- if (p_scale_factor_x < 1 && p_scale_factor_y < 1) {
- cv::resize(input_gray, input_gray, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_AREA);
- cv::resize(input_rgb, input_rgb, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_AREA);
- } else {
- cv::resize(input_gray, input_gray, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_LINEAR);
- cv::resize(input_rgb, input_rgb, cv::Size(0, 0), p_scale_factor_x, p_scale_factor_y, cv::INTER_LINEAR);
- }
}
+}
- double max_response = -1.;
- int scale_index = 0;
- cv::Point2i *max_response_pt = nullptr;
- cv::Mat *max_response_map = nullptr;
-
- if(m_use_multithreading) {
- std::vector<std::future<void>> async_res(p_scales.size());
- for (size_t i = 0; i < p_scale_vars.size(); ++i) {
- async_res[i] = std::async(std::launch::async,
- [this, &input_gray, &input_rgb, i]() -> void
- {return scale_track(this->p_scale_vars[i], input_rgb, input_gray, this->p_scales[i]);});
+double KCF_Tracker::findMaxReponse(uint &max_idx, cv::Point2d &new_location) const
+{
+ double max = -1.;
+#ifndef BIG_BATCH
+ for (uint j = 0; j < d.threadctxs.size(); ++j) {
+ if (d.threadctxs[j].max.response > max) {
+ max = d.threadctxs[j].max.response;
+ max_idx = j;
}
- for (size_t i = 0; i < p_scales.size(); ++i) {
- async_res[i].wait();
- if (this->p_scale_vars[i].max_response > max_response) {
- max_response = this->p_scale_vars[i].max_response;
- max_response_pt = & this->p_scale_vars[i].max_loc;
- max_response_map = & this->p_scale_vars[i].response;
- scale_index = i;
- }
+ }
+#else
+ for (uint j = 0; j < p_scales.size(); ++j) {
+ if (d.threadctxs[0].max[j].response > max) {
+ max = d.threadctxs[0].max[j].response;
+ max_idx = j;
}
- } else {
-#pragma omp parallel for schedule(dynamic)
- for (size_t i = 0; i < p_scale_vars.size(); ++i) {
- scale_track(this->p_scale_vars[i], input_rgb, input_gray, this->p_scales[i]);
-#pragma omp critical
- {
- if (this->p_scale_vars[i].max_response > max_response) {
- max_response = this->p_scale_vars[i].max_response;
- max_response_pt = & this->p_scale_vars[i].max_loc;
- max_response_map = & this->p_scale_vars[i].response;
- scale_index = i;
- }
+ }
+#endif
+
+ if (m_visual_debug) {
+ int w = 100; //feature_size.width;
+ int h = 100; //feature_size.height;
+ cv::Mat all_responses(h * p_num_scales, w * p_num_angles,
+ d.threadctxs[0].response.type(), cv::Scalar::all(0));
+ for (size_t i = 0; i < p_num_scales; ++i) {
+ for (size_t j = 0; j < p_num_angles; ++j) {
+ cv::Mat tmp = d.threadctxs[IF_BIG_BATCH(0, p_num_angles * i + j)].response.plane(IF_BIG_BATCH(p_num_angles * i + j, 0));
+ tmp = circshift(tmp, -tmp.cols/2, -tmp.rows/2);
+ cv::resize(tmp, tmp, cv::Size(w, h));
+ cv::Mat resp_roi(all_responses, cv::Rect(j * w, i * h, w, h));
+ tmp.copyTo(resp_roi);
}
}
+ cv::namedWindow("All responses", CV_WINDOW_AUTOSIZE);
+ cv::imshow("All responses", all_responses);
}
- DEBUG_PRINTM(*max_response_map);
- DEBUG_PRINT(*max_response_pt);
+ cv::Point2i &max_response_pt = IF_BIG_BATCH(d.threadctxs[0].max[max_idx].loc, d.threadctxs[max_idx].max.loc);
+ cv::Mat max_response_map = IF_BIG_BATCH(d.threadctxs[0].response.plane(max_idx), d.threadctxs[max_idx].response.plane(0));
- //sub pixel quadratic interpolation from neighbours
- if (max_response_pt->y > max_response_map->rows / 2) //wrap around to negative half-space of vertical axis
- max_response_pt->y = max_response_pt->y - max_response_map->rows;
- if (max_response_pt->x > max_response_map->cols / 2) //same for horizontal axis
- max_response_pt->x = max_response_pt->x - max_response_map->cols;
+ DEBUG_PRINTM(max_response_map);
+ DEBUG_PRINT(max_response_pt);
- cv::Point2f new_location(max_response_pt->x, max_response_pt->y);
- DEBUG_PRINT(new_location);
+ // sub pixel quadratic interpolation from neighbours
+ if (max_response_pt.y > max_response_map.rows / 2) // wrap around to negative half-space of vertical axis
+ max_response_pt.y = max_response_pt.y - max_response_map.rows;
+ if (max_response_pt.x > max_response_map.cols / 2) // same for horizontal axis
+ max_response_pt.x = max_response_pt.x - max_response_map.cols;
- if (m_use_subpixel_localization)
- new_location = sub_pixel_peak(*max_response_pt, *max_response_map);
- DEBUG_PRINT(new_location);
- p_pose.cx += p_current_scale*p_cell_size*new_location.x;
- p_pose.cy += p_current_scale*p_cell_size*new_location.y;
- if (p_fit_to_pw2) {
- if (p_pose.cx < 0) p_pose.cx = 0;
- if (p_pose.cx > (img.cols*p_scale_factor_x)-1) p_pose.cx = (img.cols*p_scale_factor_x)-1;
- if (p_pose.cy < 0) p_pose.cy = 0;
- if (p_pose.cy > (img.rows*p_scale_factor_y)-1) p_pose.cy = (img.rows*p_scale_factor_y)-1;
+ if (m_use_subpixel_localization) {
+ new_location = sub_pixel_peak(max_response_pt, max_response_map);
} else {
- if (p_pose.cx < 0) p_pose.cx = 0;
- if (p_pose.cx > img.cols-1) p_pose.cx = img.cols-1;
- if (p_pose.cy < 0) p_pose.cy = 0;
- if (p_pose.cy > img.rows-1) p_pose.cy = img.rows-1;
+ new_location = max_response_pt;
}
+ DEBUG_PRINT(new_location);
+ return max;
+}
- //sub grid scale interpolation
- double new_scale = p_scales[scale_index];
- if (m_use_subgrid_scale)
- new_scale = sub_grid_scale(scale_index);
+void KCF_Tracker::track(cv::Mat &img)
+{
+ __dbgTracer.debug = m_debug;
+ TRACE("");
- p_current_scale *= new_scale;
+ cv::Mat input_gray, input_rgb = img.clone();
+ if (img.channels() == 3) {
+ cv::cvtColor(img, input_gray, CV_BGR2GRAY);
+ input_gray.convertTo(input_gray, CV_32FC1);
+ } else
+ img.convertTo(input_gray, CV_32FC1);
- if (p_current_scale < p_min_max_scale[0])
- p_current_scale = p_min_max_scale[0];
- if (p_current_scale > p_min_max_scale[1])
- p_current_scale = p_min_max_scale[1];
- //obtain a subwindow for training at newly estimated target position
- get_features(input_rgb, input_gray, p_pose.cx, p_pose.cy, p_windows_size[0], p_windows_size[1], p_scale_vars[0], p_current_scale);
- fft.forward_window(p_scale_vars[0].patch_feats, p_scale_vars[0].xf, p_scale_vars[0].fw_all, m_use_cuda ? p_scale_vars[0].data_features_d : nullptr);
+ // don't need too large image
+ resizeImgs(input_rgb, input_gray);
+
+#ifdef ASYNC
+ for (auto &it : d.threadctxs)
+ it.async_res = std::async(std::launch::async, [this, &input_gray, &input_rgb, &it]() -> void {
+ it.track(*this, input_rgb, input_gray);
+ });
+ for (auto const &it : d.threadctxs)
+ it.async_res.wait();
+
+#else // !ASYNC
+ NORMAL_OMP_PARALLEL_FOR
+ for (uint i = 0; i < d.threadctxs.size(); ++i)
+ d.threadctxs[i].track(*this, input_rgb, input_gray);
+#endif
- //subsequent frames, interpolate model
- p_model_xf = p_model_xf * (1. - p_interp_factor) + p_scale_vars[0].xf * p_interp_factor;
+ cv::Point2d new_location;
+ uint max_idx;
+ max_response = findMaxReponse(max_idx, new_location);
- ComplexMat alphaf_num, alphaf_den;
+ new_location.x *= double(p_windows_size.width) / fit_size.width;
+ new_location.y *= double(p_windows_size.height) / fit_size.height;
- if (m_use_linearkernel) {
- ComplexMat xfconj = p_scale_vars[0].xf.conj();
- alphaf_num = xfconj.mul(p_yf);
- alphaf_den = (p_scale_vars[0].xf * xfconj);
+ p_current_center += p_current_scale * p_cell_size * new_location;
+
+ clamp2(p_current_center.x, 0.0, img.cols - 1.0);
+ clamp2(p_current_center.y, 0.0, img.rows - 1.0);
+
+ // sub grid scale interpolation
+ if (m_use_subgrid_scale) {
+ p_current_scale *= sub_grid_scale(max_idx);
} else {
- p_scale_vars[0].flag = Tracker_flags::AUTO_CORRELATION;
- //Kernel Ridge Regression, calculate alphas (in Fourier domain)
- gaussian_correlation(p_scale_vars[0], p_scale_vars[0].xf, p_scale_vars[0].xf, p_kernel_sigma, true);
-// ComplexMat alphaf = p_yf / (kf + p_lambda); //equation for fast training
-// p_model_alphaf = p_model_alphaf * (1. - p_interp_factor) + alphaf * p_interp_factor;
- alphaf_num = p_yf * p_scale_vars[0].kf;
- alphaf_den = p_scale_vars[0].kf * (p_scale_vars[0].kf + p_lambda);
+ p_current_scale *= p_scales[max_idx];
}
- p_model_alphaf_num = p_model_alphaf_num * (1. - p_interp_factor) + alphaf_num * p_interp_factor;
- p_model_alphaf_den = p_model_alphaf_den * (1. - p_interp_factor) + alphaf_den * p_interp_factor;
- p_model_alphaf = p_model_alphaf_num / p_model_alphaf_den;
+ clamp2(p_current_scale, p_min_max_scale[0], p_min_max_scale[1]);
+
+ // train at newly estimated target position
+ train(input_rgb, input_gray, p_interp_factor);
}
-void KCF_Tracker::scale_track(Scale_vars & vars, cv::Mat & input_rgb, cv::Mat & input_gray, double scale)
+void ThreadCtx::track(const KCF_Tracker &kcf, cv::Mat &input_rgb, cv::Mat &input_gray)
{
- get_features(input_rgb, input_gray, this->p_pose.cx, this->p_pose.cy, this->p_windows_size[0], this->p_windows_size[1],
- vars, this->p_current_scale * scale);
+ TRACE("");
+
+ BIG_BATCH_OMP_PARALLEL_FOR
+ for (uint i = 0; i < IF_BIG_BATCH(kcf.p_num_scales, 1); ++i)
+ {
+ kcf.get_features(input_rgb, input_gray, kcf.p_current_center.x, kcf.p_current_center.y,
+ kcf.p_windows_size.width, kcf.p_windows_size.height,
+ kcf.p_current_scale * IF_BIG_BATCH(kcf.p_scales[i], scale))
+ .copyTo(patch_feats.scale(i));
+ DEBUG_PRINT(patch_feats.scale(i));
+ }
- fft.forward_window(vars.patch_feats, vars.zf, vars.fw_all, m_use_cuda ? vars.data_features_d : nullptr);
- DEBUG_PRINTM(vars.zf);
+ kcf.fft.forward_window(patch_feats, zf, temp);
+ DEBUG_PRINTM(zf);
- if (m_use_linearkernel) {
- vars.kzf = (vars.zf.mul2(this->p_model_alphaf)).sum_over_channels();
- fft.inverse(vars.kzf, vars.response, m_use_cuda ? vars.data_i_1ch_d : nullptr);
+ if (kcf.m_use_linearkernel) {
+ kzf = zf.mul(kcf.p_model_alphaf).sum_over_channels();
} else {
- gaussian_correlation(vars, vars.zf, this->p_model_xf, this->p_kernel_sigma);
- DEBUG_PRINTM(this->p_model_alphaf);
- DEBUG_PRINTM(vars.kzf);
- DEBUG_PRINTM(this->p_model_alphaf * vars.kzf);
- vars.kzf = this->p_model_alphaf * vars.kzf;
- //TODO Add support for fft.inverse(vars) for CUFFT
- fft.inverse(vars.kzf, vars.response, m_use_cuda ? vars.data_i_1ch_d : nullptr);
+ gaussian_correlation(kzf, zf, kcf.p_model_xf, kcf.p_kernel_sigma, false, kcf);
+ DEBUG_PRINTM(kzf);
+ kzf = kzf.mul(kcf.p_model_alphaf);
}
+ kcf.fft.inverse(kzf, response);
- DEBUG_PRINTM(vars.response);
+ DEBUG_PRINTM(response);
/* target location is at the maximum response. we must take into
account the fact that, if the target doesn't move, the peak
will appear at the top-left corner, not at the center (this is
discussed in the paper). the responses wrap around cyclically. */
- double min_val;
- cv::Point2i min_loc;
- cv::minMaxLoc(vars.response, &min_val, &vars.max_val, &min_loc, &vars.max_loc);
+ double min_val, max_val;
+ cv::Point2i min_loc, max_loc;
+#ifdef BIG_BATCH
+ for (size_t i = 0; i < kcf.p_scales.size(); ++i) {
+ cv::minMaxLoc(response.plane(i), &min_val, &max_val, &min_loc, &max_loc);
+ DEBUG_PRINT(max_loc);
+ double weight = kcf.p_scales[i] < 1. ? kcf.p_scales[i] : 1. / kcf.p_scales[i];
+ max[i].response = max_val * weight;
+ max[i].loc = max_loc;
+ }
+#else
+ cv::minMaxLoc(response.plane(0), &min_val, &max_val, &min_loc, &max_loc);
- DEBUG_PRINT(vars.max_loc);
+ DEBUG_PRINT(max_loc);
+ DEBUG_PRINT(max_val);
- double weight = scale < 1. ? scale : 1./scale;
- vars.max_response = vars.max_val*weight;
+ double weight = scale < 1. ? scale : 1. / scale;
+ max.response = max_val * weight;
+ max.loc = max_loc;
+#endif
}
// ****************************************************************************
-void KCF_Tracker::get_features(cv::Mat & input_rgb, cv::Mat & input_gray, int cx, int cy, int size_x, int size_y, Scale_vars &vars, double scale)
+cv::Mat KCF_Tracker::get_features(cv::Mat &input_rgb, cv::Mat &input_gray, int cx, int cy,
+ int size_x, int size_y, double scale) const
{
- int size_x_scaled = floor(size_x*scale);
- int size_y_scaled = floor(size_y*scale);
+ cv::Size scaled = cv::Size(floor(size_x * scale), floor(size_y * scale));
- cv::Mat patch_gray = get_subwindow(input_gray, cx, cy, size_x_scaled, size_y_scaled);
- cv::Mat patch_rgb = get_subwindow(input_rgb, cx, cy, size_x_scaled, size_y_scaled);
+ cv::Mat patch_gray = get_subwindow(input_gray, cx, cy, scaled.width, scaled.height);
+ cv::Mat patch_rgb = get_subwindow(input_rgb, cx, cy, scaled.width, scaled.height);
- //resize to default size
- if (scale > 1.){
- //if we downsample use INTER_AREA interpolation
- cv::resize(patch_gray, patch_gray, cv::Size(size_x, size_y), 0., 0., cv::INTER_AREA);
- }else {
- cv::resize(patch_gray, patch_gray, cv::Size(size_x, size_y), 0., 0., cv::INTER_LINEAR);
+ // resize to default size
+ if (scaled.area() > fit_size.area()) {
+ // if we downsample use INTER_AREA interpolation
+ // note: this is just a guess - we may downsample in X and upsample in Y (or vice versa)
+ cv::resize(patch_gray, patch_gray, fit_size, 0., 0., cv::INTER_AREA);
+ } else {
+ cv::resize(patch_gray, patch_gray, fit_size, 0., 0., cv::INTER_LINEAR);
}
// get hog(Histogram of Oriented Gradients) features
- FHoG::extract(patch_gray, vars, 2, p_cell_size, 9);
+ std::vector<cv::Mat> hog_feat = FHoG::extract(patch_gray, 2, p_cell_size, 9);
- //get color rgb features (simple r,g,b channels)
+ // get color rgb features (simple r,g,b channels)
std::vector<cv::Mat> color_feat;
if ((m_use_color || m_use_cnfeat) && input_rgb.channels() == 3) {
- //resize to default size
- if (scale > 1.){
- //if we downsample use INTER_AREA interpolation
- cv::resize(patch_rgb, patch_rgb, cv::Size(size_x/p_cell_size, size_y/p_cell_size), 0., 0., cv::INTER_AREA);
- }else {
- cv::resize(patch_rgb, patch_rgb, cv::Size(size_x/p_cell_size, size_y/p_cell_size), 0., 0., cv::INTER_LINEAR);
+ // resize to default size
+ if (scaled.area() > (fit_size / p_cell_size).area()) {
+ // if we downsample use INTER_AREA interpolation
+ cv::resize(patch_rgb, patch_rgb, fit_size / p_cell_size, 0., 0., cv::INTER_AREA);
+ } else {
+ cv::resize(patch_rgb, patch_rgb, fit_size / p_cell_size, 0., 0., cv::INTER_LINEAR);
}
}
if (m_use_color && input_rgb.channels() == 3) {
- //use rgb color space
+ // use rgb color space
cv::Mat patch_rgb_norm;
patch_rgb.convertTo(patch_rgb_norm, CV_32F, 1. / 255., -0.5);
cv::Mat ch1(patch_rgb_norm.size(), CV_32FC1);
color_feat.insert(color_feat.end(), cn_feat.begin(), cn_feat.end());
}
- vars.patch_feats.insert(vars.patch_feats.end(), color_feat.begin(), color_feat.end());
- return;
+ hog_feat.insert(hog_feat.end(), color_feat.begin(), color_feat.end());
+
+ int size[] = {p_num_of_feats, feature_size.height, feature_size.width};
+ cv::Mat result(3, size, CV_32F);
+ for (uint i = 0; i < hog_feat.size(); ++i)
+ hog_feat[i].copyTo(cv::Mat(size[1], size[2], CV_32FC1, result.ptr(i)));
+
+ return result;
}
cv::Mat KCF_Tracker::gaussian_shaped_labels(double sigma, int dim1, int dim2)
int range_y[2] = {-dim2 / 2, dim2 - dim2 / 2};
int range_x[2] = {-dim1 / 2, dim1 - dim1 / 2};
- double sigma_s = sigma*sigma;
+ double sigma_s = sigma * sigma;
- for (int y = range_y[0], j = 0; y < range_y[1]; ++y, ++j){
- float * row_ptr = labels.ptr<float>(j);
- double y_s = y*y;
- for (int x = range_x[0], i = 0; x < range_x[1]; ++x, ++i){
- row_ptr[i] = std::exp(-0.5 * (y_s + x*x) / sigma_s);//-1/2*e^((y^2+x^2)/sigma^2)
+ for (int y = range_y[0], j = 0; y < range_y[1]; ++y, ++j) {
+ float *row_ptr = labels.ptr<float>(j);
+ double y_s = y * y;
+ for (int x = range_x[0], i = 0; x < range_x[1]; ++x, ++i) {
+ row_ptr[i] = std::exp(-0.5 * (y_s + x * x) / sigma_s); //-1/2*e^((y^2+x^2)/sigma^2)
}
}
- //rotate so that 1 is at top-left corner (see KCF paper for explanation)
-#ifdef CUFFT
- cv::Mat tmp = circshift(labels, range_x[0], range_y[0]);
- tmp.copyTo(p_scale_vars[0].rot_labels);
-
- assert(p_scale_vars[0].rot_labels.at<float>(0,0) >= 1.f - 1e-10f);
- return tmp;
-#else
- cv::Mat rot_labels = circshift(labels, range_x[0], range_y[0]);
- //sanity check, 1 at top left corner
- assert(rot_labels.at<float>(0,0) >= 1.f - 1e-10f);
+ // rotate so that 1 is at top-left corner (see KCF paper for explanation)
+ MatDynMem rot_labels = circshift(labels, range_x[0], range_y[0]);
+ // sanity check, 1 at top left corner
+ assert(rot_labels.at<float>(0, 0) >= 1.f - 1e-10f);
return rot_labels;
-#endif
}
-cv::Mat KCF_Tracker::circshift(const cv::Mat &patch, int x_rot, int y_rot)
+cv::Mat KCF_Tracker::circshift(const cv::Mat &patch, int x_rot, int y_rot) const
{
cv::Mat rot_patch(patch.size(), CV_32FC1);
cv::Mat tmp_x_rot(patch.size(), CV_32FC1);
- //circular rotate x-axis
+ // circular rotate x-axis
if (x_rot < 0) {
- //move part that does not rotate over the edge
+ // move part that does not rotate over the edge
cv::Range orig_range(-x_rot, patch.cols);
cv::Range rot_range(0, patch.cols - (-x_rot));
patch(cv::Range::all(), orig_range).copyTo(tmp_x_rot(cv::Range::all(), rot_range));
- //rotated part
+ // rotated part
orig_range = cv::Range(0, -x_rot);
rot_range = cv::Range(patch.cols - (-x_rot), patch.cols);
patch(cv::Range::all(), orig_range).copyTo(tmp_x_rot(cv::Range::all(), rot_range));
- }else if (x_rot > 0){
- //move part that does not rotate over the edge
+ } else if (x_rot > 0) {
+ // move part that does not rotate over the edge
cv::Range orig_range(0, patch.cols - x_rot);
cv::Range rot_range(x_rot, patch.cols);
patch(cv::Range::all(), orig_range).copyTo(tmp_x_rot(cv::Range::all(), rot_range));
- //rotated part
+ // rotated part
orig_range = cv::Range(patch.cols - x_rot, patch.cols);
rot_range = cv::Range(0, x_rot);
patch(cv::Range::all(), orig_range).copyTo(tmp_x_rot(cv::Range::all(), rot_range));
- }else { //zero rotation
- //move part that does not rotate over the edge
+ } else { // zero rotation
+ // move part that does not rotate over the edge
cv::Range orig_range(0, patch.cols);
cv::Range rot_range(0, patch.cols);
patch(cv::Range::all(), orig_range).copyTo(tmp_x_rot(cv::Range::all(), rot_range));
}
- //circular rotate y-axis
+ // circular rotate y-axis
if (y_rot < 0) {
- //move part that does not rotate over the edge
+ // move part that does not rotate over the edge
cv::Range orig_range(-y_rot, patch.rows);
cv::Range rot_range(0, patch.rows - (-y_rot));
tmp_x_rot(orig_range, cv::Range::all()).copyTo(rot_patch(rot_range, cv::Range::all()));
- //rotated part
+ // rotated part
orig_range = cv::Range(0, -y_rot);
rot_range = cv::Range(patch.rows - (-y_rot), patch.rows);
tmp_x_rot(orig_range, cv::Range::all()).copyTo(rot_patch(rot_range, cv::Range::all()));
- }else if (y_rot > 0){
- //move part that does not rotate over the edge
+ } else if (y_rot > 0) {
+ // move part that does not rotate over the edge
cv::Range orig_range(0, patch.rows - y_rot);
cv::Range rot_range(y_rot, patch.rows);
tmp_x_rot(orig_range, cv::Range::all()).copyTo(rot_patch(rot_range, cv::Range::all()));
- //rotated part
+ // rotated part
orig_range = cv::Range(patch.rows - y_rot, patch.rows);
rot_range = cv::Range(0, y_rot);
tmp_x_rot(orig_range, cv::Range::all()).copyTo(rot_patch(rot_range, cv::Range::all()));
- }else { //zero rotation
- //move part that does not rotate over the edge
+ } else { // zero rotation
+ // move part that does not rotate over the edge
cv::Range orig_range(0, patch.rows);
cv::Range rot_range(0, patch.rows);
tmp_x_rot(orig_range, cv::Range::all()).copyTo(rot_patch(rot_range, cv::Range::all()));
return rot_patch;
}
-//hann window actually (Power-of-cosine windows)
+// hann window actually (Power-of-cosine windows)
cv::Mat KCF_Tracker::cosine_window_function(int dim1, int dim2)
{
cv::Mat m1(1, dim1, CV_32FC1), m2(dim2, 1, CV_32FC1);
- double N_inv = 1./(static_cast<double>(dim1)-1.);
+ double N_inv = 1. / (static_cast<double>(dim1) - 1.);
for (int i = 0; i < dim1; ++i)
- m1.at<float>(i) = 0.5*(1. - std::cos(2. * CV_PI * static_cast<double>(i) * N_inv));
- N_inv = 1./(static_cast<double>(dim2)-1.);
+ m1.at<float>(i) = float(0.5 * (1. - std::cos(2. * CV_PI * static_cast<double>(i) * N_inv)));
+ N_inv = 1. / (static_cast<double>(dim2) - 1.);
for (int i = 0; i < dim2; ++i)
- m2.at<float>(i) = 0.5*(1. - std::cos(2. * CV_PI * static_cast<double>(i) * N_inv));
- cv::Mat ret = m2*m1;
+ m2.at<float>(i) = float(0.5 * (1. - std::cos(2. * CV_PI * static_cast<double>(i) * N_inv)));
+ cv::Mat ret = m2 * m1;
return ret;
}
// Returns sub-window of image input centered at [cx, cy] coordinates),
// with size [width, height]. If any pixels are outside of the image,
// they will replicate the values at the borders.
-cv::Mat KCF_Tracker::get_subwindow(const cv::Mat & input, int cx, int cy, int width, int height)
+cv::Mat KCF_Tracker::get_subwindow(const cv::Mat &input, int cx, int cy, int width, int height) const
{
cv::Mat patch;
- int x1 = cx - width/2;
- int y1 = cy - height/2;
- int x2 = cx + width/2;
- int y2 = cy + height/2;
+ int x1 = cx - width / 2;
+ int y1 = cy - height / 2;
+ int x2 = cx + width / 2;
+ int y2 = cy + height / 2;
- //out of image
+ // out of image
if (x1 >= input.cols || y1 >= input.rows || x2 < 0 || y2 < 0) {
patch.create(height, width, input.type());
- patch.setTo(0.f);
+ patch.setTo(double(0.f));
return patch;
}
int top = 0, bottom = 0, left = 0, right = 0;
- //fit to image coordinates, set border extensions;
+ // fit to image coordinates, set border extensions;
if (x1 < 0) {
left = -x1;
x1 = 0;
if (x2 - x1 == 0 || y2 - y1 == 0)
patch = cv::Mat::zeros(height, width, CV_32FC1);
- else
- {
- cv::copyMakeBorder(input(cv::Range(y1, y2), cv::Range(x1, x2)), patch, top, bottom, left, right, cv::BORDER_REPLICATE);
-// imshow( "copyMakeBorder", patch);
-// cv::waitKey();
- }
+ else {
+ cv::copyMakeBorder(input(cv::Range(y1, y2), cv::Range(x1, x2)), patch, top, bottom, left, right,
+ cv::BORDER_REPLICATE);
+ // imshow( "copyMakeBorder", patch);
+ // cv::waitKey();
+ }
- //sanity check
+ // sanity check
assert(patch.cols == width && patch.rows == height);
return patch;
}
-void KCF_Tracker::gaussian_correlation(struct Scale_vars & vars, const ComplexMat & xf, const ComplexMat & yf, double sigma, bool auto_correlation)
+void KCF_Tracker::GaussianCorrelation::operator()(ComplexMat &result, const ComplexMat &xf, const ComplexMat &yf,
+ double sigma, bool auto_correlation, const KCF_Tracker &kcf)
{
-#ifdef CUFFT
- xf.sqr_norm(vars.xf_sqr_norm_d);
- if (!auto_correlation)
- yf.sqr_norm(vars.yf_sqr_norm_d);
-#else
- xf.sqr_norm(vars.xf_sqr_norm);
- if (auto_correlation){
- vars.yf_sqr_norm[0] = vars.xf_sqr_norm[0];
+ TRACE("");
+ xf.sqr_norm(xf_sqr_norm);
+ if (auto_correlation) {
+ yf_sqr_norm = xf_sqr_norm;
} else {
- yf.sqr_norm(vars.yf_sqr_norm);
+ yf.sqr_norm(yf_sqr_norm);
}
-#endif
- vars.xyf = auto_correlation ? xf.sqr_mag() : xf.mul2(yf.conj());
- DEBUG_PRINTM(vars.xyf);
- fft.inverse(vars.xyf, vars.ifft2_res, m_use_cuda ? vars.data_i_features_d : nullptr);
+ xyf = auto_correlation ? xf.sqr_mag() : xf * yf.conj(); // xf.muln(yf.conj());
+ DEBUG_PRINTM(xyf);
+
+ // ifft2 and sum over 3rd dimension, we dont care about individual channels
+ ComplexMat xyf_sum = xyf.sum_over_channels();
+ DEBUG_PRINTM(xyf_sum);
+ kcf.fft.inverse(xyf_sum, ifft_res);
+ DEBUG_PRINTM(ifft_res);
#ifdef CUFFT
- if(auto_correlation)
- cuda_gaussian_correlation(vars.data_i_features, vars.gauss_corr_res_d, vars.xf_sqr_norm_d, vars.xf_sqr_norm_d, sigma, xf.n_channels, xf.n_scales, p_roi_height, p_roi_width);
- else
- cuda_gaussian_correlation(vars.data_i_features, vars.gauss_corr_res_d, vars.xf_sqr_norm_d, vars.yf_sqr_norm_d, sigma, xf.n_channels, xf.n_scales, p_roi_height, p_roi_width);
+ // FIXME
+ cuda_gaussian_correlation(ifft_res.deviceMem(), k.deviceMem(), xf_sqr_norm.deviceMem(),
+ auto_correlation ? xf_sqr_norm.deviceMem() : yf_sqr_norm.deviceMem(), sigma,
+ xf.n_channels, xf.n_scales, kcf.feature_size.height, kcf.feature_size.width);
#else
- //ifft2 and sum over 3rd dimension, we dont care about individual channels
- DEBUG_PRINTM(vars.ifft2_res);
- cv::Mat xy_sum;
- if (xf.channels() != p_num_scales*p_num_of_feats)
- xy_sum.create(vars.ifft2_res.size(), CV_32FC1);
- else
- xy_sum.create(vars.ifft2_res.size(), CV_32FC(p_scales.size()));
- xy_sum.setTo(0);
- for (int y = 0; y < vars.ifft2_res.rows; ++y) {
- float * row_ptr = vars.ifft2_res.ptr<float>(y);
- float * row_ptr_sum = xy_sum.ptr<float>(y);
- for (int x = 0; x < vars.ifft2_res.cols; ++x) {
- for (int sum_ch = 0; sum_ch < xy_sum.channels(); ++sum_ch) {
- row_ptr_sum[(x*xy_sum.channels())+sum_ch] += std::accumulate(row_ptr + x*vars.ifft2_res.channels() + sum_ch*(vars.ifft2_res.channels()/xy_sum.channels()),
- (row_ptr + x*vars.ifft2_res.channels() + (sum_ch+1)*(vars.ifft2_res.channels()/xy_sum.channels())), 0.f);
- }
- }
- }
- DEBUG_PRINTM(xy_sum);
- std::vector<cv::Mat> scales;
- cv::split(xy_sum,scales);
-
- float numel_xf_inv = 1.f/(xf.cols * xf.rows * (xf.channels()/xf.n_scales));
- for (int i = 0; i < xf.n_scales; ++i){
- cv::Mat in_roi(vars.in_all, cv::Rect(0, i*scales[0].rows, scales[0].cols, scales[0].rows));
- cv::exp(- 1.f / (sigma * sigma) * cv::max((vars.xf_sqr_norm[i] + vars.yf_sqr_norm[0] - 2 * scales[i]) * numel_xf_inv, 0), in_roi);
- DEBUG_PRINTM(in_roi);
+ float numel_xf_inv = 1.f / (xf.cols * xf.rows * (xf.channels() / xf.n_scales));
+ for (uint i = 0; i < xf.n_scales; ++i) {
+ cv::Mat plane = ifft_res.plane(i);
+ DEBUG_PRINT(ifft_res.plane(i));
+ cv::exp(-1. / (sigma * sigma) * cv::max((xf_sqr_norm[i] + yf_sqr_norm[0] - 2 * ifft_res.plane(i))
+ * numel_xf_inv, 0), plane);
+ DEBUG_PRINTM(plane);
}
#endif
- DEBUG_PRINTM(vars.in_all);
- fft.forward(vars.in_all, auto_correlation ? vars.kf : vars.kzf, m_use_cuda ? vars.gauss_corr_res_d : nullptr);
- return;
+ kcf.fft.forward(ifft_res, result);
}
-float get_response_circular(cv::Point2i & pt, cv::Mat & response)
+float get_response_circular(cv::Point2i &pt, cv::Mat &response)
{
int x = pt.x;
int y = pt.y;
- if (x < 0)
- x = response.cols + x;
- if (y < 0)
- y = response.rows + y;
- if (x >= response.cols)
- x = x - response.cols;
- if (y >= response.rows)
- y = y - response.rows;
-
- return response.at<float>(y,x);
+ assert(response.dims == 2); // ensure .cols and .rows are valid
+ if (x < 0) x = response.cols + x;
+ if (y < 0) y = response.rows + y;
+ if (x >= response.cols) x = x - response.cols;
+ if (y >= response.rows) y = y - response.rows;
+
+ return response.at<float>(y, x);
}
-cv::Point2f KCF_Tracker::sub_pixel_peak(cv::Point & max_loc, cv::Mat & response)
+cv::Point2f KCF_Tracker::sub_pixel_peak(cv::Point &max_loc, cv::Mat &response) const
{
- //find neighbourhood of max_loc (response is circular)
+ // find neighbourhood of max_loc (response is circular)
// 1 2 3
// 4 5
// 6 7 8
- cv::Point2i p1(max_loc.x-1, max_loc.y-1), p2(max_loc.x, max_loc.y-1), p3(max_loc.x+1, max_loc.y-1);
- cv::Point2i p4(max_loc.x-1, max_loc.y), p5(max_loc.x+1, max_loc.y);
- cv::Point2i p6(max_loc.x-1, max_loc.y+1), p7(max_loc.x, max_loc.y+1), p8(max_loc.x+1, max_loc.y+1);
+ cv::Point2i p1(max_loc.x - 1, max_loc.y - 1), p2(max_loc.x, max_loc.y - 1), p3(max_loc.x + 1, max_loc.y - 1);
+ cv::Point2i p4(max_loc.x - 1, max_loc.y), p5(max_loc.x + 1, max_loc.y);
+ cv::Point2i p6(max_loc.x - 1, max_loc.y + 1), p7(max_loc.x, max_loc.y + 1), p8(max_loc.x + 1, max_loc.y + 1);
// clang-format off
// fit 2d quadratic function f(x, y) = a*x^2 + b*x*y + c*y^2 + d*x + e*y + f
cv::Mat x;
cv::solve(A, fval, x, cv::DECOMP_SVD);
- double a = x.at<float>(0), b = x.at<float>(1), c = x.at<float>(2),
- d = x.at<float>(3), e = x.at<float>(4);
+ float a = x.at<float>(0), b = x.at<float>(1), c = x.at<float>(2), d = x.at<float>(3), e = x.at<float>(4);
cv::Point2f sub_peak(max_loc.x, max_loc.y);
if (b > 0 || b < 0) {
return sub_peak;
}
-double KCF_Tracker::sub_grid_scale(int index)
+double KCF_Tracker::sub_grid_scale(uint index)
{
cv::Mat A, fval;
- if (index < 0 || index > (int)p_scales.size()-1) {
+ if (index >= p_scales.size()) {
// interpolate from all values
// fit 1d quadratic function f(x) = a*x^2 + b*x + c
A.create(p_scales.size(), 3, CV_32FC1);
fval.create(p_scales.size(), 1, CV_32FC1);
for (size_t i = 0; i < p_scales.size(); ++i) {
- A.at<float>(i, 0) = p_scales[i] * p_scales[i];
- A.at<float>(i, 1) = p_scales[i];
+ A.at<float>(i, 0) = float(p_scales[i] * p_scales[i]);
+ A.at<float>(i, 1) = float(p_scales[i]);
A.at<float>(i, 2) = 1;
- fval.at<float>(i) = p_scale_vars[i].max_response;
+ fval.at<float>(i) = d.threadctxs.back().IF_BIG_BATCH(max[i].response, max.response);
}
} else {
- //only from neighbours
- if (index == 0 || index == (int)p_scales.size()-1)
- return p_scales[index];
+ // only from neighbours
+ if (index == 0 || index == p_scales.size() - 1)
+ return p_scales[index];
A = (cv::Mat_<float>(3, 3) <<
- p_scales[index-1] * p_scales[index-1], p_scales[index-1], 1,
- p_scales[index] * p_scales[index], p_scales[index], 1,
- p_scales[index+1] * p_scales[index+1], p_scales[index+1], 1);
- fval = (cv::Mat_<float>(3, 1) << p_scale_vars[index-1].max_response, p_scale_vars[index].max_response, p_scale_vars[index+1].max_response);
+ p_scales[index - 1] * p_scales[index - 1], p_scales[index - 1], 1,
+ p_scales[index + 0] * p_scales[index + 0], p_scales[index + 0], 1,
+ p_scales[index + 1] * p_scales[index + 1], p_scales[index + 1], 1);
+#ifdef BIG_BATCH
+ fval = (cv::Mat_<float>(3, 1) <<
+ d.threadctxs.back().max[index - 1].response,
+ d.threadctxs.back().max[index + 0].response,
+ d.threadctxs.back().max[index + 1].response);
+#else
+ fval = (cv::Mat_<float>(3, 1) <<
+ d.threadctxs[index - 1].max.response,
+ d.threadctxs[index + 0].max.response,
+ d.threadctxs[index + 1].max.response);
+#endif
}
cv::Mat x;
cv::solve(A, fval, x, cv::DECOMP_SVD);
- double a = x.at<float>(0), b = x.at<float>(1);
+ float a = x.at<float>(0), b = x.at<float>(1);
double scale = p_scales[index];
if (a > 0 || a < 0)
scale = -b / (2 * a);
projects / hercules2020 / kcf.git / blobdiff