run darknet inference using libdarknet.so

darknet_inference.cpp

//g++ -std=c++11 libdarknet.so test.cpp `pkg-config --libs --cflags opencv
#include "include/darknet.h"
#include <iostream>
#include <vector>
#include <opencv2/opencv.hpp>
#include <opencv2/core.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/highgui.hpp>
using namespace std;
using namespace cv;

struct mybbox{
    int best_class;
    float prob;
    float x;
    float y;
    float w;
    float h;
    mybbox(int cls, float prob, float x, float y, float w, float h){
        this->best_class = cls;
        this->prob = prob;
        this->x = x;
        this->y = y;
        this->w = w;
        this->h = h;
    }
};

int main(){
    cv::Mat img = cv::imread("video_19_530.jpg");
    network net = *load_network_custom("./signate_yolov4_tiny/yolov4-tiny-custom.cfg", "./signate_yolov4_tiny/yolov4-tiny-custom_best.weights", 0, 1);
    image im = load_image_color("video_19_530.jpg", 0, 0);
    image sized = resize_image(im, net.w, net.h);
    int nboxes = 0;
    int letter_box = 0;
    float thresh = 0.5f;
    float hier_thresh = 0.5f;

    network_predict(net, sized.data);

    detection *dets = get_network_boxes(&net, im.w, im.h, thresh, hier_thresh, 0, 1, &nboxes, letter_box);
    float nms = 0.4f;
    layer l = net.layers[net.n - 1];
    do_nms_sort(dets, nboxes, l.classes, nms);
    //remove negatives (cf darknet.py)
    cout << "classses:" << l.classes << endl;
    vector<mybbox> final_results;
    for(int i = 0; i < nboxes; i++){
        detection det = dets[i];
        for(int j = 0; j < l.classes; j++){
            if(det.prob[j] > 0){
                final_results.push_back(mybbox(j, det.prob[j], det.bbox.x, det.bbox.y, det.bbox.w, det.bbox.h));
            }
        }
    }

    cout << "nboxes " << nboxes << " final bboxes " << final_results.size() << endl;

    for(mybbox box : final_results){
        cout << box.best_class << " " << box.x << " " << box.y << " " << box.w << " " << box.h << " " << box.prob <<  endl;
        float xmin = (box.x - box.w / 2) * img.cols;
        float xmax = (box.x + box.w / 2) * img.cols;
        float ymin = (box.y - box.h / 2) * img.rows;
        float ymax = (box.y + box.h / 2) * img.rows;

        rectangle(img, Point(xmin, ymin), Point(xmax, ymax),
              Scalar(0, 255, 0), 3, 1, 0);
    }
    cv::imwrite("myresult.jpg", img);
    free_detections(dets, nboxes);
    free_image(im);
    free_image(sized);
    free_network(net);

}