main_vot.cpp

   1 #include <stdlib.h>
   2 #include <getopt.h>
   3 #include <libgen.h>
   4 #include <unistd.h>
   5
   6 #include "kcf.h"
   7 #include "vot.hpp"
   8
   9 double calcAccuracy(std::string line, cv::Rect bb_rect, cv::Rect &groundtruth_rect)
  10 {
  11     std::vector<float> numbers;
  12     std::istringstream s( line );
  13     float x;
  14     char ch;
  15
  16     while (s >> x){
  17         numbers.push_back(x);
  18         s >> ch;
  19     }
  20     double x1 = std::min(numbers[0], std::min(numbers[2], std::min(numbers[4], numbers[6])));
  21     double x2 = std::max(numbers[0], std::max(numbers[2], std::max(numbers[4], numbers[6])));
  22     double y1 = std::min(numbers[1], std::min(numbers[3], std::min(numbers[5], numbers[7])));
  23     double y2 = std::max(numbers[1], std::max(numbers[3], std::max(numbers[5], numbers[7])));
  24
  25     groundtruth_rect = cv::Rect(x1, y1, x2-x1, y2-y1);
  26
  27     double rects_intersection = (groundtruth_rect & bb_rect).area();
  28     double rects_union = (groundtruth_rect | bb_rect).area();
  29     double accuracy = rects_intersection/rects_union;
  30
  31     return accuracy;
  32 }
  33
  34 int main(int argc, char *argv[])
  35 {
  36     //load region, images and prepare for output
  37     std::string region, images, output;
  38     int visualize_delay = -1, fit_size_x = -1, fit_size_y = -1;
  39     KCF_Tracker tracker;
  40
  41     while (1) {
  42         int option_index = 0;
  43         static struct option long_options[] = {
  44             {"debug",     no_argument,       0,  'd' },
  45             {"help",      no_argument,       0,  'h' },
  46             {"output",    required_argument, 0,  'o' },
  47             {"visualize", optional_argument, 0,  'v' },
  48             {"fit",       optional_argument, 0,  'f' },
  49             {0,           0,                 0,  0 }
  50         };
  51
  52         int c = getopt_long(argc, argv, "dhv::f::o:",
  53                         long_options, &option_index);
  54         if (c == -1)
  55             break;
  56
  57         switch (c) {
  58         case 'd':
  59             tracker.m_debug = true;
  60             break;
  61         case 'h':
  62             std::cerr << "Usage: \n"
  63                       << argv[0] << " [options]\n"
  64                       << argv[0] << " [options] <directory>\n"
  65                       << argv[0] << " [options] <path/to/region.txt or groundtruth.txt> <path/to/images.txt> [path/to/output.txt]\n"
  66                       << "Options:\n"
  67                       << " --visualize | -v[delay_ms]\n"
  68                       << " --output    | -o <output.txt>\n"
  69                       << " --debug     | -d\n"
  70                       << " --fit       | -f[WxH]\n";
  71             exit(0);
  72             break;
  73         case 'o':
  74             output = optarg;
  75             break;
  76         case 'v':
  77             visualize_delay = optarg ? atol(optarg) : 1;
  78             break;
  79         case 'f':
  80             std::string sizes = optarg ? optarg : "128x128";
  81             std::string delimiter = "x";
  82             size_t pos = sizes.find(delimiter);
  83             std::string first_argument = sizes.substr(0, pos);
  84             sizes.erase(0, pos + delimiter.length());
  85
  86             fit_size_x = stol(first_argument);
  87             fit_size_y = stol(sizes);
  88             break;
  89         }
  90     }
  91
  92     switch (argc - optind) {
  93     case 1:
  94         if (chdir(argv[optind]) == -1) {
  95             perror(argv[optind]);
  96             exit(1);
  97         }
  98         // Fall through
  99     case 0:
 100         region = access("groundtruth.txt", F_OK) == 0 ? "groundtruth.txt" : "region.txt";
 101         images = "images.txt";
 102         if (output.empty())
 103             output = "output.txt";
 104         break;
 105     case 2:
 106         // Fall through
 107     case 3:
 108         region = std::string(argv[optind + 0]);
 109         images = std::string(argv[optind + 1]);
 110         if (output.empty()) {
 111             if ((argc - optind) == 3)
 112                 output = std::string(argv[optind + 2]);
 113             else
 114                 output = std::string(dirname(argv[optind + 0])) + "/output.txt";
 115         }
 116         break;
 117     default:
 118         std::cerr << "Too many arguments\n";
 119         return 1;
 120     }
 121     VOT vot_io(region, images, output);
 122
 123     // if groundtruth.txt is used use intersection over union (IOU) to calculate tracker accuracy
 124     std::ifstream groundtruth_stream;
 125     if (region.compare("groundtruth.txt") == 0) {
 126         groundtruth_stream.open(region.c_str());
 127         std::string line;
 128         std::getline(groundtruth_stream, line);
 129     }
 130
 131     cv::Mat image;
 132
 133     //img = firts frame, initPos = initial position in the first frame
 134     cv::Rect init_rect = vot_io.getInitRectangle();
 135     vot_io.outputBoundingBox(init_rect);
 136     vot_io.getNextImage(image);
 137
 138     tracker.init(image, init_rect, fit_size_x, fit_size_y);
 139
 140     BBox_c bb;
 141     cv::Rect bb_rect;
 142     double avg_time = 0., sum_accuracy = 0.;
 143     int frames = 0;
 144     while (vot_io.getNextImage(image) == 1){
 145         double time_profile_counter = cv::getCPUTickCount();
 146         tracker.track(image);
 147         time_profile_counter = cv::getCPUTickCount() - time_profile_counter;
 148          std::cout << "  -> speed : " <<  time_profile_counter/((double)cvGetTickFrequency()*1000) << "ms. per frame, "
 149                       "response : " << tracker.getFilterResponse();
 150         avg_time += time_profile_counter/((double)cvGetTickFrequency()*1000);
 151         frames++;
 152
 153         bb = tracker.getBBox();
 154         bb_rect = cv::Rect(bb.cx - bb.w/2., bb.cy - bb.h/2., bb.w, bb.h);
 155         vot_io.outputBoundingBox(bb_rect);
 156
 157         if (groundtruth_stream.is_open()) {
 158             std::string line;
 159             std::getline(groundtruth_stream, line);
 160
 161             cv::Rect groundtruthRect;
 162             double accuracy = calcAccuracy(line, bb_rect, groundtruthRect);
 163             if (visualize_delay >= 0)
 164                 cv::rectangle(image, groundtruthRect, CV_RGB(255, 0,0), 1);
 165             std::cout << ", accuracy: " << accuracy;
 166             sum_accuracy += accuracy;
 167         }
 168
 169         std::cout << std::endl;
 170
 171         if (visualize_delay >= 0) {
 172             cv::rectangle(image, bb_rect, CV_RGB(0,255,0), 2);
 173             cv::imshow("output", image);
 174             int ret = cv::waitKey(visualize_delay);
 175             if (visualize_delay > 0 && ret != -1 && ret != 255)
 176                 break;
 177         }
 178
 179 //        std::stringstream s;
 180 //        std::string ss;
 181 //        int countTmp = frames;
 182 //        s << "imgs" << "/img" << (countTmp/10000);
 183 //        countTmp = countTmp%10000;
 184 //        s << (countTmp/1000);
 185 //        countTmp = countTmp%1000;
 186 //        s << (countTmp/100);
 187 //        countTmp = countTmp%100;
 188 //        s << (countTmp/10);
 189 //        countTmp = countTmp%10;
 190 //        s << (countTmp);
 191 //        s << ".jpg";
 192 //        s >> ss;
 193 //        //set image output parameters
 194 //        std::vector<int> compression_params;
 195 //        compression_params.push_back(CV_IMWRITE_JPEG_QUALITY);
 196 //        compression_params.push_back(90);
 197 //        cv::imwrite(ss.c_str(), image, compression_params);
 198     }
 199
 200     std::cout << "Average processing speed: " << avg_time/frames <<  "ms (" << 1./(avg_time/frames)*1000 << " fps)";
 201     if (groundtruth_stream.is_open()) {
 202         std::cout << "; Average accuracy: " << sum_accuracy/frames << std::endl;
 203         groundtruth_stream.close();
 204     }
 205     std::cout << std::endl;
 206
 207     return EXIT_SUCCESS;
 208 }