#include <thread>
#include <algorithm>
#include "threadctx.hpp"
-#include <ios>
-#include <stdarg.h>
-#include <stdio.h>
+#include "debug.h"
#ifdef FFTW
#include "fft_fftw.h"
#include <omp.h>
#endif // OPENMP
-class DbgTracer {
- int indentLvl = 0;
-
- public:
- bool debug = false;
-
- std::string indent() { return std::string(indentLvl * 4, ' '); }
-
- class FTrace {
- DbgTracer &t;
- const char *funcName;
-
- public:
- FTrace(DbgTracer &dt, const char *fn, const char *format, ...) : t(dt), funcName(fn)
- {
- if (!t.debug) return;
- char *arg;
- va_list vl;
- va_start(vl, format);
- if (-1 == vasprintf(&arg, format, vl))
- throw std::runtime_error("vasprintf error");
- va_end(vl);
-
- std::cerr << t.indent() << funcName << "(" << arg << ") {" << std::endl;
- dt.indentLvl++;
- }
- ~FTrace()
- {
- if (!t.debug) return;
- t.indentLvl--;
- std::cerr << t.indent() << "}" << std::endl;
- }
- };
-
- template <typename T>
- void traceVal(const char *name, const T& obj, int line)
- {
- (void)line;
- if (debug)
- std::cerr << indent() << name /*<< " @" << line */ << " " << print(obj) << std::endl;
- }
-
- template <typename T> struct Printer {
- const T &obj;
- Printer(const T &_obj) : obj(_obj) {}
- };
-
- template <typename T> Printer<T> print(const T& obj) { return Printer<T>(obj); }
- Printer<cv::Mat> print(const MatScales& obj) { return Printer<cv::Mat>(obj); }
- Printer<cv::Mat> print(const MatFeats& obj) { return Printer<cv::Mat>(obj); }
- Printer<cv::Mat> print(const MatScaleFeats& obj) { return Printer<cv::Mat>(obj); }
-};
-
-template <typename T>
-std::ostream &operator<<(std::ostream &os, const DbgTracer::Printer<T> &p) {
- os << p.obj;
- return os;
-}
-std::ostream &operator<<(std::ostream &os, const DbgTracer::Printer<cv::Mat> &p) {
- os << p.obj.size << " " << p.obj.channels() << "ch " << static_cast<const void*>(p.obj.data);
- os << " = [ ";
- constexpr size_t num = 10;
- for (size_t i = 0; i < std::min(num, p.obj.total()); ++i)
- os << *p.obj.ptr<float>(i) << ", ";
- os << (num < p.obj.total() ? "... ]" : "]");
- return os;
-}
-#if defined(CUFFT)
-std::ostream &operator<<(std::ostream &os, const cufftComplex &p) {
- (void)p; // TODO
- return os;
-}
-#endif
-template <>
-std::ostream &operator<<(std::ostream &os, const DbgTracer::Printer<ComplexMat> &p) {
- os << "<cplx> " << p.obj.size() << " " << p.obj.channels() << "ch " << p.obj.get_p_data();
- os << " = [ ";
- constexpr int num = 10;
- for (int i = 0; i < std::min(num, p.obj.size().area()); ++i)
- os << p.obj.get_p_data()[i] << ", ";
- os << (num < p.obj.size().area() ? "... ]" : "]");
- return os;
-}
-
DbgTracer __dbgTracer;
-#define TRACE(...) const DbgTracer::FTrace __tracer(__dbgTracer, __PRETTY_FUNCTION__, ##__VA_ARGS__)
-
-#define DEBUG_PRINT(obj) __dbgTracer.traceVal(#obj, (obj), __LINE__)
-#define DEBUG_PRINTM(obj) DEBUG_PRINT(obj)
-
-
template <typename T>
T clamp(const T& n, const T& lower, const T& upper)
{
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), d(*new Kcf_Tracker_Private)
+ : 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)
{
}
delete &d;
}
-void KCF_Tracker::train(cv::Mat input_gray, cv::Mat input_rgb, double interp_factor)
+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, p_roi);
+ DEBUG_PRINT(patch_feats);
MatScaleFeats temp(1, p_num_of_feats, p_roi);
get_features(input_rgb, input_gray, p_pose.cx, p_pose.cy,
p_windows_size.width, p_windows_size.height,
- p_current_scale).copyTo(patch_feats.features(0));
-
+ 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);
alphaf_den = (p_xf * xfconj);
} else {
// Kernel Ridge Regression, calculate alphas (in Fourier domain)
- const uint num_scales = BIG_BATCH_MODE ? p_num_scales : 1;
cv::Size sz(Fft::freq_size(p_roi));
- ComplexMat kf(sz.height, sz.width, num_scales);
- (*gaussian_correlation)(*this, kf, p_model_xf, p_model_xf, p_kernel_sigma, true);
+ 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;
- DEBUG_PRINTM(p_model_alphaf_num);
p_model_alphaf_den = kf * (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_roi.height = p_windows_size.height / p_cell_size;
p_scales.clear();
- if (m_use_scale)
- 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));
- 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_roi.height * (p_roi.width / 2 + 1) > 1024) {
std::cerr << "cuFFT supports only Gaussian kernel." << std::endl;
std::exit(EXIT_FAILURE);
}
- CudaSafeCall(cudaSetDeviceFlags(cudaDeviceMapHost));
#else
p_xf.create(p_roi.height, p_roi.height / 2 + 1, p_num_of_feats);
#endif
d.threadctxs.emplace_back(p_roi, p_num_of_feats, p_num_scales);
#endif
- gaussian_correlation.reset(
- new GaussianCorrelation(IF_BIG_BATCH(p_num_scales, 1), p_roi));
+ gaussian_correlation.reset(new GaussianCorrelation(1, p_roi));
p_current_scale = 1.;
DEBUG_PRINTM(p_yf);
// train initial model
- train(input_gray, input_rgb, 1.0);
+ 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)
tmp.w *= p_current_scale;
tmp.h *= p_current_scale;
- if (p_resize_image) tmp.scale(1 / p_downscale_factor);
+ 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);
}
}
-void KCF_Tracker::findMaxReponse(uint &max_idx, cv::Point2f &new_location) const
+double KCF_Tracker::findMaxReponse(uint &max_idx, cv::Point2f &new_location) const
{
double max = -1.;
#ifndef BIG_BATCH
}
}
#else
- // FIXME: Iterate correctly in big batch mode - perhaps have only one element in the list
for (uint j = 0; j < p_scales.size(); ++j) {
if (d.threadctxs[0].max[j].response > max) {
max = d.threadctxs[0].max[j].response;
new_location = max_response_pt;
}
DEBUG_PRINT(new_location);
+ return max;
}
void KCF_Tracker::track(cv::Mat &img)
it.async_res.wait();
#else // !ASYNC
- // FIXME: Iterate correctly in big batch mode - perhaps have only one element in the list
NORMAL_OMP_PARALLEL_FOR
for (uint i = 0; i < d.threadctxs.size(); ++i)
d.threadctxs[i].track(*this, input_rgb, input_gray);
cv::Point2f new_location;
uint max_idx;
- findMaxReponse(max_idx, new_location);
+ max_response = findMaxReponse(max_idx, new_location);
p_pose.cx += p_current_scale * p_cell_size * double(new_location.x);
p_pose.cy += p_current_scale * p_cell_size * double(new_location.y);
kcf.get_features(input_rgb, input_gray, kcf.p_pose.cx, kcf.p_pose.cy,
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.features(i));
- DEBUG_PRINT(patch_feats.features(i));
+ .copyTo(patch_feats.scale(i));
+ DEBUG_PRINT(patch_feats.scale(i));
}
- DEBUG_PRINT(patch_feats);
kcf.fft.forward_window(patch_feats, zf, temp);
DEBUG_PRINTM(zf);
if (kcf.m_use_linearkernel) {
kzf = zf.mul(kcf.p_model_alphaf).sum_over_channels();
} else {
- gaussian_correlation(kcf, kzf, zf, kcf.p_model_xf, kcf.p_kernel_sigma);
- DEBUG_PRINTM(kcf.p_model_alphaf);
+ gaussian_correlation(kzf, zf, kcf.p_model_xf, kcf.p_kernel_sigma, false, kcf);
DEBUG_PRINTM(kzf);
kzf = kzf.mul(kcf.p_model_alphaf);
- DEBUG_PRINTM(kzf);
}
kcf.fft.inverse(kzf, response);
cv::minMaxLoc(response.plane(0), &min_val, &max_val, &min_loc, &max_loc);
DEBUG_PRINT(max_loc);
+ DEBUG_PRINT(max_val);
double weight = scale < 1. ? scale : 1. / scale;
max.response = max_val * weight;
return patch;
}
-void KCF_Tracker::GaussianCorrelation::operator()(const KCF_Tracker &kcf, ComplexMat &result, 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)
{
TRACE("");
xf.sqr_norm(xf_sqr_norm);
yf.sqr_norm(yf_sqr_norm);
}
xyf = auto_correlation ? xf.sqr_mag() : xf * yf.conj(); // xf.muln(yf.conj());
- //DEBUG_PRINTM(xyf);
- kcf.fft.inverse(xyf, ifft_res);
+ 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
+ // 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.p_roi.height, kcf.p_roi.width);
#else
- // ifft2 and sum over 3rd dimension, we dont care about individual channels
- //DEBUG_PRINTM(ifft_res);
- cv::Mat xy_sum;
- if (xf.channels() != kcf.p_num_scales * kcf.p_num_of_feats)
- xy_sum.create(ifft_res.size(), CV_32FC1);
- else
- xy_sum.create(ifft_res.size(), CV_32FC(kcf.p_scales.size()));
- xy_sum.setTo(0);
- for (int y = 0; y < ifft_res.rows; ++y) {
- float *row_ptr = ifft_res.ptr<float>(y);
- float *row_ptr_sum = xy_sum.ptr<float>(y);
- for (int x = 0; x < ifft_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 * ifft_res.channels() + sum_ch * (ifft_res.channels() / xy_sum.channels()),
- (row_ptr + x * ifft_res.channels() +
- (sum_ch + 1) * (ifft_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 (uint i = 0; i < xf.n_scales; ++i) {
- cv::Mat k_roi = k.plane(i);
- cv::exp(-1. / (sigma * sigma) * cv::max((xf_sqr_norm[i] + yf_sqr_norm[0] - 2 * scales[i]) * numel_xf_inv, 0),
- k_roi);
- DEBUG_PRINTM(k_roi);
+ 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
- kcf.fft.forward(k, result);
+ kcf.fft.forward(ifft_res, result);
}
float get_response_circular(cv::Point2i &pt, cv::Mat &response)
{
int x = pt.x;
int y = pt.y;
+ 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;