main.cpp

#define _CRT_SECURE_NO_WARNINGS
#include <fstream>
#include <iostream>
#include <opencv2/dnn.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/highgui.hpp>
#include <assert.h>

using namespace cv;
using namespace dnn;
using namespace std;

class NanoDet
{
	public:
		NanoDet(int input_shape, float confThreshold, float nmsThreshold);
		void detect(Mat& srcimg);

	private:
		const int stride[3] = { 8, 16, 32 };
		const string classesFile = "coco.names";   ////这个是存放COCO数据集的类名，如果你是用自己数据集训练的，那么需要修改
		int input_shape[2];   //// height, width
		const float mean[3] = { 103.53, 116.28, 123.675 };
		const float std[3] = { 57.375, 57.12, 58.395 };
		const int reg_max = 7;
		float prob_threshold;
		float iou_threshold;
        vector<string> classes;
        int num_class;
		Net net;

        Mat resize_image(Mat srcimg, int* newh, int* neww, int* top, int* left);
        void normalize(Mat& srcimg);
        void softmax(float* x, int length);
        void post_process(vector<Mat> outs, Mat& frame, int newh, int neww, int top, int left);
        void generate_proposal(vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, const int stride_, Mat out_score, Mat out_box);
		const bool keep_ratio = true;
};

NanoDet::NanoDet(int input_shape, float confThreshold, float nmsThreshold)
{
    assert(input_shape==320 || input_shape==416);
    this->input_shape[0] = input_shape;
    this->input_shape[1] = input_shape;
	this->prob_threshold = confThreshold;
	this->iou_threshold = nmsThreshold;

	ifstream ifs(this->classesFile.c_str());
	string line;
	while (getline(ifs, line)) this->classes.push_back(line);
    this->num_class = this->classes.size();
    if(input_shape==320)
    {
        this->net = readNet("nanodet.onnx");
    }
	else
    {
        this->net = readNet("nanodet_m.onnx");
    }
}

Mat NanoDet::resize_image(Mat srcimg, int* newh, int* neww, int* top, int* left)
{
	int srch = srcimg.rows, srcw = srcimg.cols;
	*newh = this->input_shape[0];
	*neww = this->input_shape[1];
	Mat dstimg;
	if (this->keep_ratio && srch != srcw)
	{
		float hw_scale = (float)srch / srcw;
		if (hw_scale > 1)
		{
			*newh = this->input_shape[0];
			*neww = int(this->input_shape[1] / hw_scale);
			resize(srcimg, dstimg, Size(*neww, *newh), INTER_AREA);
			*left = int((this->input_shape[1] - *neww) * 0.5);
			copyMakeBorder(dstimg, dstimg, 0, 0, *left, this->input_shape[1] - *neww - *left, BORDER_CONSTANT, 0);
		}
		else
		{
			*newh = (int)this->input_shape[0] * hw_scale;
			*neww = this->input_shape[1];
			resize(srcimg, dstimg, Size(*neww, *newh), INTER_AREA);
			*top = (int)(this->input_shape[0] - *newh) * 0.5;
			copyMakeBorder(dstimg, dstimg, *top, this->input_shape[0] - *newh - *top, 0, 0, BORDER_CONSTANT, 0);
		}
	}
	else
	{
		resize(srcimg, dstimg, Size(*neww, *newh), INTER_AREA);
	}
	return dstimg;
}

void NanoDet::normalize(Mat& img)
{
    img.convertTo(img, CV_32F);
	int i = 0, j = 0;
	for (i = 0; i < img.rows; i++)
	{
		float* pdata = (float*)(img.data + i * img.step);
		for (j = 0; j < img.cols; j++)
		{
			pdata[0] = (pdata[0] - this->mean[0]) / this->std[0];
			pdata[1] = (pdata[1] - this->mean[1]) / this->std[1];
			pdata[2] = (pdata[2] - this->mean[2]) / this->std[2];
			pdata += 3;
		}

//        float* pdata = img.ptr<float>(i);
//        for(j = 0; j < img.cols; j++)
//        {
//            pdata[3 * j] = (pdata[3 * j] - this->mean[0]) / this->std[0];
//            pdata[3 * j + 1] = (pdata[3 * j + 1] - this->mean[1]) / this->std[1];
//            pdata[3 * j + 2] = (pdata[3 * j + 2] - this->mean[2]) / this->std[2];
//        }
	}
}

//Mat NanoDet::normalize(Mat src)
//{
//	vector<Mat> bgrChannels(3);
//	split(src, bgrChannels);
//	for (auto i = 0; i < bgrChannels.size(); i++)
//	{
//		bgrChannels[i].convertTo(bgrChannels[i], CV_32FC1, 1.0 / this->std[i], (0.0 - this->mean[i]) / this->std[i]);
//	}
//	Mat dst;
//	merge(bgrChannels, dst);
//	return dst;
//}

void NanoDet::detect(Mat& srcimg)
{
	int newh = 0, neww = 0, top = 0, left = 0;
	Mat dstimg = this->resize_image(srcimg, &newh, &neww, &top, &left);
	this->normalize(dstimg);
    Mat blob = blobFromImage(dstimg);

	this->net.setInput(blob);
	vector<Mat> outs;
	this->net.forward(outs, this->net.getUnconnectedOutLayersNames());
	this->post_process(outs, srcimg, newh, neww, top, left);
}

void NanoDet::softmax(float* x, int length)
{
	float sum = 0;
	int i = 0;
	for (i = 0; i < length; i++)
	{
		x[i] = exp(x[i]);
		sum += x[i];
	}
	for (i = 0; i < length; i++)
	{
		x[i] /= sum;
	}
}

void NanoDet::generate_proposal(vector<int>& classIds, vector<float>& confidences, vector<Rect>& boxes, const int stride_, Mat out_score, Mat out_box)
{
    const int num_grid_y = (int)this->input_shape[0]/stride_;
    const int num_grid_x = (int)this->input_shape[1]/stride_;
    const int reg_1max = this->reg_max + 1;

    if(out_score.dims==3)
    {
        out_score = out_score.reshape(0, num_grid_x*num_grid_y);
    }
    if(out_box.dims==3)
    {
        out_box = out_box.reshape(0, num_grid_x*num_grid_y);
    }
    for (int i = 0; i < num_grid_y; i++)
    {
        for (int j = 0; j < num_grid_x; j++)
        {
            const int idx = i * num_grid_x + j;
			Mat scores = out_score.row(idx).colRange(0, num_class);
			Point classIdPoint;
			double score;
			// Get the value and location of the maximum score
			minMaxLoc(scores, 0, &score, 0, &classIdPoint);
            if (score >= this->prob_threshold)
            {
                float* pbox = (float*)out_box.data + idx * reg_1max * 4;
                float dis_pred[4];
                for (int k = 0; k < 4; k++)
                {
                    this->softmax(pbox, reg_1max);
                    float dis = 0.f;
                    for (int l = 0; l < reg_1max; l++)
                    {
                        dis += l * pbox[l];
                    }
                    dis_pred[k] = dis * stride_;
                    pbox += reg_1max;
                }

				float pb_cx = (j + 0.5f) * stride_ - 0.5;
                float pb_cy = (i + 0.5f) * stride_ - 0.5;       
                float x0 = pb_cx - dis_pred[0];
                float y0 = pb_cy - dis_pred[1];
                float x1 = pb_cx + dis_pred[2];
                float y1 = pb_cy + dis_pred[3];

                classIds.push_back(classIdPoint.x);
                confidences.push_back(score);
                boxes.push_back(Rect((int)x0, (int)y0, (int)(x1 - x0), (int)(y1 - y0)));
            }
        }
    }
}

void NanoDet::post_process(vector<Mat> outs, Mat& frame, int newh, int neww, int top, int left)
{
    /////generate proposals
    vector<int> classIds;
    vector<float> confidences;
    vector<Rect> boxes;
    this->generate_proposal(classIds, confidences, boxes, this->stride[0], outs[0], outs[1]);
    this->generate_proposal(classIds, confidences, boxes, this->stride[1], outs[2], outs[3]);
    this->generate_proposal(classIds, confidences, boxes, this->stride[2], outs[4], outs[5]);

    // Perform non maximum suppression to eliminate redundant overlapping boxes with
    // lower confidences
    vector<int> indices;
    NMSBoxes(boxes, confidences, this->prob_threshold, this->iou_threshold, indices);
    float ratioh = (float)frame.rows / newh;
    float ratiow = (float)frame.cols / neww;
    for (size_t i = 0; i < indices.size(); ++i)
    {
        int idx = indices[i];
        Rect box = boxes[idx];
        int xmin = (int)max((box.x - left)*ratiow, 0.f);
        int ymin = (int)max((box.y - top)*ratioh, 0.f);
        int xmax = (int)min((box.x - left + box.width)*ratiow, (float)frame.cols);
        int ymax = (int)min((box.y - top + box.height)*ratioh, (float)frame.rows);
        rectangle(frame, Point(xmin, ymin), Point(xmax, ymax), Scalar(0, 0, 255), 3);
        //Get the label for the class name and its confidence
        string label = format("%.2f", confidences[idx]);
        label = classes[classIds[idx]] + ":" + label;

        //Display the label at the top of the bounding box
        int baseLine;
        Size labelSize = getTextSize(label, FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);
        ymin = max(ymin, labelSize.height);
        //rectangle(frame, Point(left, top - int(1.5 * labelSize.height)), Point(left + int(1.5 * labelSize.width), top + baseLine), Scalar(0, 255, 0), FILLED);
        putText(frame, label, Point(xmin, ymin), FONT_HERSHEY_SIMPLEX, 0.75, Scalar(0, 255, 0), 1);
    }
}

int main()
{
	NanoDet nanonet(416, 0.35, 0.6);

	string imgpath = "street.png";  ///输入图片的路径，你也可以改成外部传参argv的方式，或者是读取视频文件
	Mat srcimg = imread(imgpath);
	nanonet.detect(srcimg);

	static const string kWinName = "Deep learning object detection in OpenCV";
	namedWindow(kWinName, WINDOW_NORMAL);
	imshow(kWinName, srcimg);
	waitKey(0);
	destroyAllWindows();
}