10 double calcAccuracy(std::string line, cv::Rect bb_rect, cv::Rect &groundtruth_rect)
12 std::vector<float> numbers;
13 std::istringstream s(line);
21 double x1 = std::min(numbers[0], std::min(numbers[2], std::min(numbers[4], numbers[6])));
22 double x2 = std::max(numbers[0], std::max(numbers[2], std::max(numbers[4], numbers[6])));
23 double y1 = std::min(numbers[1], std::min(numbers[3], std::min(numbers[5], numbers[7])));
24 double y2 = std::max(numbers[1], std::max(numbers[3], std::max(numbers[5], numbers[7])));
26 groundtruth_rect = cv::Rect(x1, y1, x2 - x1, y2 - y1);
28 double rects_intersection = (groundtruth_rect & bb_rect).area();
29 double rects_union = (groundtruth_rect | bb_rect).area();
30 double accuracy = rects_intersection / rects_union;
35 int main(int argc, char *argv[])
37 //load region, images and prepare for output
38 std::string region, images, output;
39 int visualize_delay = -1, fit_size_x = -1, fit_size_y = -1;
44 static struct option long_options[] = {
45 {"debug", no_argument, 0, 'd' },
46 {"visualDebug", no_argument, 0, 'p'},
47 {"help", no_argument, 0, 'h' },
48 {"output", required_argument, 0, 'o' },
49 {"visualize", optional_argument, 0, 'v' },
50 {"fit", optional_argument, 0, 'f' },
54 int c = getopt_long(argc, argv, "dphv::f::o:", long_options, &option_index);
60 tracker.m_debug = true;
63 tracker.m_visual_debug = true;
64 visualize_delay = 500;
67 std::cerr << "Usage: \n"
68 << argv[0] << " [options]\n"
69 << argv[0] << " [options] <directory>\n"
70 << argv[0] << " [options] <path/to/region.txt or groundtruth.txt> <path/to/images.txt> [path/to/output.txt]\n"
72 << " --visualize | -v[delay_ms]\n"
73 << " --output | -o <output.txt>\n"
75 << " --fit | -f[W[xH]]\n";
82 visualize_delay = optarg ? atol(optarg) : 1;
86 fit_size_x = fit_size_y = 0;
89 if (sscanf(optarg, "%d%c", &fit_size_x, &tail) == 1) {
90 fit_size_y = fit_size_x;
91 } else if (sscanf(optarg, "%dx%d%c", &fit_size_x, &fit_size_y, &tail) != 2) {
92 fprintf(stderr, "Cannot parse -f argument: %s\n", optarg);
100 switch (argc - optind) {
102 if (chdir(argv[optind]) == -1) {
103 perror(argv[optind]);
108 region = access("groundtruth.txt", F_OK) == 0 ? "groundtruth.txt" : "region.txt";
109 images = "images.txt";
111 output = "output.txt";
116 region = std::string(argv[optind + 0]);
117 images = std::string(argv[optind + 1]);
118 if (output.empty()) {
119 if ((argc - optind) == 3)
120 output = std::string(argv[optind + 2]);
122 output = std::string(dirname(argv[optind + 0])) + "/output.txt";
126 std::cerr << "Too many arguments\n";
129 VOT vot_io(region, images, output);
131 // if groundtruth.txt is used use intersection over union (IOU) to calculate tracker accuracy
132 std::ifstream groundtruth_stream;
133 if (region.compare("groundtruth.txt") == 0) {
134 groundtruth_stream.open(region.c_str());
136 std::getline(groundtruth_stream, line);
141 //img = firts frame, initPos = initial position in the first frame
142 cv::Rect init_rect = vot_io.getInitRectangle();
143 vot_io.outputBoundingBox(init_rect);
144 vot_io.getNextImage(image);
146 tracker.init(image, init_rect, fit_size_x, fit_size_y);
150 double avg_time = 0., sum_accuracy = 0.;
153 std::cout << std::fixed << std::setprecision(2);
155 while (vot_io.getNextImage(image) == 1){
156 double time_profile_counter = cv::getCPUTickCount();
157 tracker.track(image);
158 time_profile_counter = cv::getCPUTickCount() - time_profile_counter;
159 std::cout << " -> speed : " << time_profile_counter/((double)cvGetTickFrequency()*1000) << "ms per frame, "
160 "response : " << tracker.getFilterResponse();
161 avg_time += time_profile_counter/((double)cvGetTickFrequency()*1000);
164 bb = tracker.getBBox();
165 bb_rect = cv::Rect(bb.cx - bb.w/2., bb.cy - bb.h/2., bb.w, bb.h);
166 vot_io.outputBoundingBox(bb_rect);
168 if (groundtruth_stream.is_open()) {
170 std::getline(groundtruth_stream, line);
172 cv::Rect groundtruthRect;
173 double accuracy = calcAccuracy(line, bb_rect, groundtruthRect);
174 if (visualize_delay >= 0)
175 cv::rectangle(image, groundtruthRect, CV_RGB(255, 0,0), 1);
176 std::cout << ", accuracy: " << accuracy;
177 sum_accuracy += accuracy;
180 std::cout << std::endl;
182 if (visualize_delay >= 0) {
183 cv::rectangle(image, bb_rect, CV_RGB(0,255,0), 2);
184 cv::imshow("output", image);
185 int ret = cv::waitKey(visualize_delay);
186 if (visualize_delay > 0 && ret != -1 && ret < 128)
190 // std::stringstream s;
192 // int countTmp = frames;
193 // s << "imgs" << "/img" << (countTmp/10000);
194 // countTmp = countTmp%10000;
195 // s << (countTmp/1000);
196 // countTmp = countTmp%1000;
197 // s << (countTmp/100);
198 // countTmp = countTmp%100;
199 // s << (countTmp/10);
200 // countTmp = countTmp%10;
204 // //set image output parameters
205 // std::vector<int> compression_params;
206 // compression_params.push_back(CV_IMWRITE_JPEG_QUALITY);
207 // compression_params.push_back(90);
208 // cv::imwrite(ss.c_str(), image, compression_params);
211 std::cout << "Average processing speed: " << avg_time / frames << "ms (" << 1. / (avg_time / frames) * 1000 << " fps)";
212 if (groundtruth_stream.is_open()) {
213 std::cout << "; Average accuracy: " << sum_accuracy/frames << std::endl;
214 groundtruth_stream.close();
216 std::cout << std::endl;