使用OpenCV+ONNXRuntime部署YOLOV7目标检测

简单说明

分别使用OpenCV、ONNXRuntime部署YOLOV7目标检测，一共包含12个onnx模型，依然是包含C++和Python两个版本的程序。 编写这套YOLOV7的程序，跟此前编写的YOLOV6的程序，大部分源码是相同的，区别仅仅在于图片预处理的过程不一样。YOLOV7的图片预处理是BGR2RGB+不保持高宽比的resize+除以255 由于onnx文件太多，无法直接上传到仓库里，需要从百度云盘下载，

下载完成后把models目录放在主程序文件的目录内，编译运行 使用opencv部署的程序，有一个待优化的问题。onnxruntime读取.onnx文件可以获得输入张量的形状信息， 但是opencv的dnn模块读取.onnx文件无法获得输入张量的形状信息，目前是根据.onnx文件的名称来解析字符串获得输入张量的高度和宽度的。

YOLOV7的训练源码是：

跟YOLOR是同一个作者的。

OpenCV+YOLOv7

推理过程跟之前的YOLO系列部署代码可以大部分重用！这里就不在赘述了，详细看源码如下：输出部分直接解析最后一个输出层就好啦！

详细实现代码如下：

#include
#include
#include
#include
#include
#include

usingnamespacecv;
usingnamespacednn;
usingnamespacestd;

structNet_config
{
floatconfThreshold;//Confidencethreshold
floatnmsThreshold;//Non-maximumsuppressionthreshold
stringmodelpath;
};

classYOLOV7
{
public:
YOLOV7(Net_configconfig);
voiddetect(Mat&frame);
private:
intinpWidth;
intinpHeight;
vectorclass_names;
intnum_class;

floatconfThreshold;
floatnmsThreshold;
Netnet;
voiddrawPred(floatconf,intleft,inttop,intright,intbottom,Mat&frame,intclassid);
};

YOLOV7::YOLOV7(Net_configconfig)
{
this->confThreshold=config.confThreshold;
this->nmsThreshold=config.nmsThreshold;

this->net=readNet(config.modelpath);
ifstreamifs("coco.names");
stringline;
while(getline(ifs,line))this->class_names.push_back(line);
this->num_class=class_names.size();

size_tpos=config.modelpath.find("_");
intlen=config.modelpath.length()-6-pos;
stringhxw=config.modelpath.substr(pos+1,len);
pos=hxw.find("x");
stringh=hxw.substr(0,pos);
len=hxw.length()-pos;
stringw=hxw.substr(pos+1,len);
this->inpHeight=stoi(h);
this->inpWidth=stoi(w);
}

voidYOLOV7::drawPred(floatconf,intleft,inttop,intright,intbottom,Mat&frame,intclassid)//Drawthepredictedboundingbox
{
//Drawarectangledisplayingtheboundingbox
rectangle(frame,Point(left,top),Point(right,bottom),Scalar(0,0,255),2);

//Getthelabelfortheclassnameanditsconfidence
stringlabel=format("%.2f",conf);
label=this->class_names[classid]+":"+label;

//Displaythelabelatthetopoftheboundingbox
intbaseLine;
SizelabelSize=getTextSize(label,FONT_HERSHEY_SIMPLEX,0.5,1,&baseLine);
top=max(top,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(left,top),FONT_HERSHEY_SIMPLEX,0.75,Scalar(0,255,0),1);
}

voidYOLOV7::detect(Mat&frame)
{
Matblob=blobFromImage(frame,1/255.0,Size(this->inpWidth,this->inpHeight),Scalar(0,0,0),true,false);
this->net.setInput(blob);
vectorouts;
this->net.forward(outs,this->net.getUnconnectedOutLayersNames());

intnum_proposal=outs[0].size[0];
intnout=outs[0].size[1];
if(outs[0].dims>2)
{
num_proposal=outs[0].size[1];
nout=outs[0].size[2];
outs[0]=outs[0].reshape(0,num_proposal);
}
/////generateproposals
vectorconfidences;
vectorboxes;
vectorclassIds;
floatratioh=(float)frame.rows/this->inpHeight,ratiow=(float)frame.cols/this->inpWidth;
intn=0,row_ind=0;///cx,cy,w,h,box_score,class_score
float*pdata=(float*)outs[0].data;
for(n=0;n< num_proposal; n++)   ///ÌØÕ÷ͼ³ß¶È
    {
        float box_score = pdata[4];
        if (box_score >this->confThreshold)
{
Matscores=outs[0].row(row_ind).colRange(5,nout);
PointclassIdPoint;
doublemax_class_socre;
//Getthevalueandlocationofthemaximumscore
minMaxLoc(scores,0,&max_class_socre,0,&classIdPoint);
max_class_socre*=box_score;
if(max_class_socre>this->confThreshold)
{
constintclass_idx=classIdPoint.x;
floatcx=pdata[0]*ratiow;///cx
floatcy=pdata[1]*ratioh;///cy
floatw=pdata[2]*ratiow;///w
floath=pdata[3]*ratioh;///h

intleft=int(cx-0.5*w);
inttop=int(cy-0.5*h);

confidences.push_back((float)max_class_socre);
boxes.push_back(Rect(left,top,(int)(w),(int)(h)));
classIds.push_back(class_idx);
}
}
row_ind++;
pdata+=nout;
}

//Performnonmaximumsuppressiontoeliminateredundantoverlappingboxeswith
//lowerconfidences
vectorindices;
dnn::NMSBoxes(boxes,confidences,this->confThreshold,this->nmsThreshold,indices);
for(size_ti=0;i< indices.size(); ++i)
    {
        int idx = indices[i];
        Rect box = boxes[idx];
        this->drawPred(confidences[idx],box.x,box.y,
box.x+box.width,box.y+box.height,frame,classIds[idx]);
}
}

intmain()
{
Net_configYOLOV7_nets={0.3,0.5,"models/yolov7_736x1280.onnx"};////choices=["models/yolov7_736x1280.onnx","models/yolov7-tiny_384x640.onnx","models/yolov7_480x640.onnx","models/yolov7_384x640.onnx","models/yolov7-tiny_256x480.onnx","models/yolov7-tiny_256x320.onnx","models/yolov7_256x320.onnx","models/yolov7-tiny_256x640.onnx","models/yolov7_256x640.onnx","models/yolov7-tiny_480x640.onnx","models/yolov7-tiny_736x1280.onnx","models/yolov7_256x480.onnx"]
YOLOV7net(YOLOV7_nets);
stringimgpath="images/dog.jpg";
Matsrcimg=imread(imgpath);
net.detect(srcimg);

staticconststringkWinName="DeeplearningobjectdetectioninOpenCV";
namedWindow(kWinName,WINDOW_NORMAL);
imshow(kWinName,srcimg);
waitKey(0);
destroyAllWindows();
}

运行测试如下：

审核编辑：刘清