| | |
| | | image images[FRAMES]; |
| | | float *avg; |
| | | float *predictions[FRAMES]; |
| | | int demo_index; |
| | | }; |
| | | |
| | | |
| | | YOLODLL_API Detector::Detector(std::string cfg_filename, std::string weight_filename, int gpu_id) |
| | | { |
| | | int old_gpu_index; |
| | | #ifdef GPU |
| | | cudaGetDevice(&old_gpu_index); |
| | | #endif |
| | | |
| | | detector_gpu_ptr = std::make_shared<detector_gpu_t>(); |
| | | detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get()); |
| | | |
| | | #ifdef GPU |
| | | cudaSetDevice(gpu_id); |
| | | #endif |
| | | network &net = detector_gpu.net; |
| | | net.gpu_index = gpu_id; |
| | | //gpu_index = i; |
| | |
| | | detector_gpu.probs = (float **)calloc(l.w*l.h*l.n, sizeof(float *)); |
| | | for (j = 0; j < l.w*l.h*l.n; ++j) detector_gpu.probs[j] = (float *)calloc(l.classes, sizeof(float)); |
| | | |
| | | #ifdef GPU |
| | | cudaSetDevice(old_gpu_index); |
| | | #endif |
| | | } |
| | | |
| | | |
| | |
| | | for (int j = 0; j < FRAMES; ++j) free(detector_gpu.predictions[j]); |
| | | for (int j = 0; j < FRAMES; ++j) if(detector_gpu.images[j].data) free(detector_gpu.images[j].data); |
| | | |
| | | for (int j = 0; j < l.w*l.h*l.n; ++j) free(detector_gpu.probs[j]); |
| | | free(detector_gpu.boxes); |
| | | free(detector_gpu.probs); |
| | | for (int j = 0; j < l.w*l.h*l.n; ++j) free(detector_gpu.probs[j]); |
| | | |
| | | int old_gpu_index; |
| | | #ifdef GPU |
| | | cudaGetDevice(&old_gpu_index); |
| | | cudaSetDevice(detector_gpu.net.gpu_index); |
| | | #endif |
| | | |
| | | free_network(detector_gpu.net); |
| | | |
| | | #ifdef GPU |
| | | cudaSetDevice(old_gpu_index); |
| | | #endif |
| | | } |
| | | |
| | | YOLODLL_API int Detector::get_net_width() { |
| | | detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get()); |
| | | return detector_gpu.net.w; |
| | | } |
| | | YOLODLL_API int Detector::get_net_height() { |
| | | detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get()); |
| | | return detector_gpu.net.h; |
| | | } |
| | | |
| | | |
| | | YOLODLL_API std::vector<bbox_t> Detector::detect(std::string image_filename, float thresh) |
| | | YOLODLL_API std::vector<bbox_t> Detector::detect(std::string image_filename, float thresh, bool use_mean) |
| | | { |
| | | std::shared_ptr<image_t> image_ptr(new image_t, [](image_t *img) { if (img->data) free(img->data); delete img; }); |
| | | *image_ptr = load_image(image_filename); |
| | | return detect(*image_ptr, thresh); |
| | | return detect(*image_ptr, thresh, use_mean); |
| | | } |
| | | |
| | | static image load_image_stb(char *filename, int channels) |
| | |
| | | } |
| | | } |
| | | |
| | | YOLODLL_API std::vector<bbox_t> Detector::detect(image_t img, float thresh) |
| | | YOLODLL_API std::vector<bbox_t> Detector::detect(image_t img, float thresh, bool use_mean) |
| | | { |
| | | |
| | | detector_gpu_t &detector_gpu = *reinterpret_cast<detector_gpu_t *>(detector_gpu_ptr.get()); |
| | | network &net = detector_gpu.net; |
| | | int old_gpu_index; |
| | | #ifdef GPU |
| | | cudaGetDevice(&old_gpu_index); |
| | | cudaSetDevice(net.gpu_index); |
| | | #endif |
| | | //std::cout << "net.gpu_index = " << net.gpu_index << std::endl; |
| | | |
| | | float nms = .4; |
| | | //float nms = .4; |
| | | |
| | | image im; |
| | | im.c = img.c; |
| | |
| | | im.h = img.h; |
| | | im.w = img.w; |
| | | |
| | | image sized = resize_image(im, net.w, net.h); |
| | | image sized; |
| | | |
| | | if (net.w == im.w && net.h == im.h) { |
| | | sized = make_image(im.w, im.h, im.c); |
| | | memcpy(sized.data, im.data, im.w*im.h*im.c * sizeof(float)); |
| | | } |
| | | else |
| | | sized = resize_image(im, net.w, net.h); |
| | | |
| | | layer l = net.layers[net.n - 1]; |
| | | |
| | | float *X = sized.data; |
| | | |
| | | network_predict(net, X); |
| | | float *prediction = network_predict(net, X); |
| | | |
| | | if (use_mean) { |
| | | memcpy(detector_gpu.predictions[detector_gpu.demo_index], prediction, l.outputs * sizeof(float)); |
| | | mean_arrays(detector_gpu.predictions, FRAMES, l.outputs, detector_gpu.avg); |
| | | l.output = detector_gpu.avg; |
| | | detector_gpu.demo_index = (detector_gpu.demo_index + 1) % FRAMES; |
| | | } |
| | | |
| | | get_region_boxes(l, 1, 1, thresh, detector_gpu.probs, detector_gpu.boxes, 0, 0); |
| | | if (nms) do_nms_sort(detector_gpu.boxes, detector_gpu.probs, l.w*l.h*l.n, l.classes, nms); |
| | |
| | | bbox.h = b.h*im.h; |
| | | bbox.obj_id = obj_id; |
| | | bbox.prob = prob; |
| | | bbox.track_id = 0; |
| | | |
| | | bbox_vec.push_back(bbox); |
| | | } |
| | |
| | | if(sized.data) |
| | | free(sized.data); |
| | | |
| | | #ifdef GPU |
| | | cudaSetDevice(old_gpu_index); |
| | | #endif |
| | | |
| | | return bbox_vec; |
| | | } |
| | | |
| | | YOLODLL_API std::vector<bbox_t> Detector::tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story) |
| | | { |
| | | bool prev_track_id_present = false; |
| | | for (auto &i : prev_bbox_vec_deque) |
| | | if (i.size() > 0) prev_track_id_present = true; |
| | | |
| | | static unsigned int track_id = 1; |
| | | |
| | | if (!prev_track_id_present) { |
| | | //track_id = 1; |
| | | for (size_t i = 0; i < cur_bbox_vec.size(); ++i) |
| | | cur_bbox_vec[i].track_id = track_id++; |
| | | prev_bbox_vec_deque.push_front(cur_bbox_vec); |
| | | if (prev_bbox_vec_deque.size() > frames_story) prev_bbox_vec_deque.pop_back(); |
| | | return cur_bbox_vec; |
| | | } |
| | | |
| | | std::vector<unsigned int> dist_vec(cur_bbox_vec.size(), std::numeric_limits<unsigned int>::max()); |
| | | |
| | | for (auto &prev_bbox_vec : prev_bbox_vec_deque) { |
| | | for (auto &i : prev_bbox_vec) { |
| | | int cur_index = -1; |
| | | for (size_t m = 0; m < cur_bbox_vec.size(); ++m) { |
| | | bbox_t const& k = cur_bbox_vec[m]; |
| | | if (i.obj_id == k.obj_id) { |
| | | unsigned int cur_dist = sqrt(((float)i.x - k.x)*((float)i.x - k.x) + ((float)i.y - k.y)*((float)i.y - k.y)); |
| | | if (cur_dist < 100 && (k.track_id == 0 || dist_vec[m] > cur_dist)) { |
| | | dist_vec[m] = cur_dist; |
| | | cur_index = m; |
| | | } |
| | | } |
| | | } |
| | | |
| | | bool track_id_absent = !std::any_of(cur_bbox_vec.begin(), cur_bbox_vec.end(), [&](bbox_t const& b) { return b.track_id == i.track_id; }); |
| | | |
| | | if (cur_index >= 0 && track_id_absent) { |
| | | cur_bbox_vec[cur_index].track_id = i.track_id; |
| | | cur_bbox_vec[cur_index].w = (cur_bbox_vec[cur_index].w + i.w) / 2; |
| | | cur_bbox_vec[cur_index].h = (cur_bbox_vec[cur_index].h + i.h) / 2; |
| | | } |
| | | } |
| | | } |
| | | |
| | | for (size_t i = 0; i < cur_bbox_vec.size(); ++i) |
| | | if (cur_bbox_vec[i].track_id == 0) |
| | | cur_bbox_vec[i].track_id = track_id++; |
| | | |
| | | prev_bbox_vec_deque.push_front(cur_bbox_vec); |
| | | if (prev_bbox_vec_deque.size() > frames_story) prev_bbox_vec_deque.pop_back(); |
| | | |
| | | return cur_bbox_vec; |
| | | } |