Added detection on images from the txt list file by using SO/DLL.

AlexeyAB

2017-08-04 e34f0416f507499e9dbbc2557430850ba3a022ab

Added detection on images from the txt list file by using SO/DLL.

 src/yolo_console_dll.cpp

@@ -73,15 +73,15 @@
#ifdef OPENCV
            std::string const file_ext = filename.substr(filename.find_last_of(".") + 1);
            if (file_ext == "avi" || file_ext == "mp4" || file_ext == "mjpg" || file_ext == "mov") {    // video file
                cv::Mat frame, prev_frame;
                cv::Mat frame, prev_frame, det_frame;
                std::vector<bbox_t> result_vec, thread_result_vec;
                detector.nms = 0.02;    // comment it - if track_id is not required
                std::thread td([]() {});
                for (cv::VideoCapture cap(filename); cap >> frame, cap.isOpened();) {
                    td.join();
                    result_vec = thread_result_vec;
                    cv::Mat det_frame = frame;
                    td = std::thread([&]() { thread_result_vec = detector.detect(det_frame, 0.2); });
                    det_frame = frame;
                    td = std::thread([&]() { thread_result_vec = detector.detect(det_frame, 0.2, true); });

                    if (!prev_frame.empty()) {
                        result_vec = detector.tracking(result_vec); // comment it - if track_id is not required
@@ -91,6 +91,16 @@
                    prev_frame = frame;
                }
            }
            else if (file_ext == "txt") {   // list of image files
                std::ifstream file(filename);
                if (!file.is_open()) std::cout << "File not found! \n";
                else 
                    for (std::string line; file >> line;) {
                        std::cout << line << std::endl;
                        show_result(detector.detect(cv::imread(line)), obj_names);
                    }
				
            }
            else {  // image file
                cv::Mat mat_img = cv::imread(filename);
                std::vector<bbox_t> result_vec = detector.detect(mat_img);

 src/yolo_v2_class.cpp

@@ -29,6 +29,7 @@
    image images[FRAMES];
    float *avg;
    float *predictions[FRAMES];
    int demo_index;
};


@@ -112,11 +113,11 @@
}


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)
@@ -163,7 +164,7 @@
    }
}

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());
@@ -196,7 +197,14 @@

    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);
@@ -269,8 +277,11 @@

            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)
            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;
            }
        }
    }


 src/yolo_v2_class.hpp

@@ -47,8 +47,8 @@
    YOLODLL_API Detector(std::string cfg_filename, std::string weight_filename, int gpu_id = 0);
    YOLODLL_API ~Detector();

    YOLODLL_API std::vector<bbox_t> detect(std::string image_filename, float thresh = 0.2);
    YOLODLL_API std::vector<bbox_t> detect(image_t img, float thresh = 0.2);
    YOLODLL_API std::vector<bbox_t> detect(std::string image_filename, float thresh = 0.2, bool use_mean = false);
    YOLODLL_API std::vector<bbox_t> detect(image_t img, float thresh = 0.2, bool use_mean = false);
    static YOLODLL_API image_t load_image(std::string image_filename);
    static YOLODLL_API void free_image(image_t m);
    YOLODLL_API int get_net_width();
@@ -57,7 +57,7 @@
    YOLODLL_API std::vector<bbox_t> tracking(std::vector<bbox_t> cur_bbox_vec, int const frames_story = 4);

#ifdef OPENCV
    std::vector<bbox_t> detect(cv::Mat mat, float thresh = 0.2) 
    std::vector<bbox_t> detect(cv::Mat mat, float thresh = 0.2, bool use_mean = false)
    {
        if(mat.data == NULL)
            throw std::runtime_error("file not found");