微信公众号:OpenCV学堂
前一阵子YOLOv4发布了,后面就是YOLOv5,估计再过几天就要YOLOv10086了,这个时代技术进步太魔幻,改几个参数就可以继续升级版本。2020.718 OpenCV4.4发布了,支持YOLOv4推理,于是我立刻测试了一波。
模型下载
YOLOv4的相关模型合集在这里
代码语言:javascript复制https://github.com/AlexeyAB/darknet/wiki/YOLOv4-model-zoo
我使用的是基于COCO预训练模型:
YOLOv4-Leaky
OpenCV4.4 DNN
OpenCV4.4 支持YOLOv4,这个是它的官方release里面说的,其实我早就发现了YOLOv4可以通过OpenCV4.2直接跑,怎么OpenCV4.4才官宣。也许不发布新版本不好官宣,只有发布了新版本才可以顺便说一下。此外OpenCV4.4 DNN还有很多新添加的演示程序,支持了深度学习的光流、支持tensorflow object detection API的EfficientDet对象检测模型,但是前提是tensorflow2.x才可以。多了一个tf_text_graph_efficientdet.py文件,用来生成对应的pbtxt文件。
OpenCV4.4 DNN YOLOv4对象检测演示
跟YOLOv3一样,YOLOv4也有三个输出层,完成推理之后,需要在进一步通过NMS实现对重叠框的去除,什么是NMS(非最大抑制),看下图就懂啦:
然后说一下模型输入格式与输出格式
输入:NCHW=1x3x416x416 输出:NXC 其中N表示多少个对象,C的前四个数矩形框的[center_x, center_y, width, height],从第五个数值开始分别是每个类别的得分,求的最大得分,如果高于阈值0.5,则认为检测到了对象,每个score对应的index即是COCO类别文本。
根据上面的描述,对一个视频文件实现YOLOv4的对象检测代码如下:
代码语言:javascript复制Net net = readNetFromDarknet(yolov4_config, yolov4_model);
net.setPreferableBackend(DNN_BACKEND_INFERENCE_ENGINE);
net.setPreferableTarget(DNN_TARGET_CPU);
std::vector<String> outNames = net.getUnconnectedOutLayersNames();
for (int i = 0; i < outNames.size(); i ) {
printf("output layer name : %sn", outNames[i].c_str());
}
vector<string> classNamesVec;
ifstream classNamesFile("D:/projects/opencv_tutorial/data/models/object_detection_classes_yolov3.txt");
if (classNamesFile.is_open())
{
string className = "";
while (std::getline(classNamesFile, className))
classNamesVec.push_back(className);
}
VideoCapture capture;
capture.open("D:/images/video/f35_02.mp4");
Mat frame;
// 加载图像
while (true) {
int64 start = getTickCount();
capture.read(frame);
Mat inputBlob = blobFromImage(frame, 1 / 255.F, Size(416, 416), Scalar(), true, false);
net.setInput(inputBlob);
// 检测
std::vector<Mat> outs;
net.forward(outs, outNames);
vector<Rect> boxes;
vector<int> classIds;
vector<float> confidences;
for (size_t i = 0; i<outs.size(); i)
{
// detected objects and C is a number of classes 4 where the first 4
float* data = (float*)outs[i].data;
for (int j = 0; j < outs[i].rows; j, data = outs[i].cols)
{
Mat scores = outs[i].row(j).colRange(5, outs[i].cols);
Point classIdPoint;
double confidence;
minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
if (confidence > 0.5)
{
int centerX = (int)(data[0] * frame.cols);
int centerY = (int)(data[1] * frame.rows);
int width = (int)(data[2] * frame.cols);
int height = (int)(data[3] * frame.rows);
int left = centerX - width / 2;
int top = centerY - height / 2;
classIds.push_back(classIdPoint.x);
confidences.push_back((float)confidence);
boxes.push_back(Rect(left, top, width, height));
}
}
}
vector<int> indices;
NMSBoxes(boxes, confidences, 0.5, 0.2, indices);
for (size_t i = 0; i < indices.size(); i)
{
int idx = indices[i];
Rect box = boxes[idx];
String className = classNamesVec[classIds[idx]];
putText(frame, className.c_str(), box.tl(), FONT_HERSHEY_SIMPLEX, 1.0, Scalar(255, 0, 0), 2, 8);
rectangle(frame, box, Scalar(0, 0, 255), 2, 8, 0);
}
float fps = getTickFrequency() / (getTickCount() - start);
float time = (getTickCount() - start) / getTickFrequency();
ostringstream ss;
ss << "FPS : "<< fps <<" detection time: " << time*1000 << " ms";
putText(frame, ss.str(), Point(20, 20), 0, 0.5, Scalar(0, 0, 255));
imshow("YOLOv4-Detections", frame);
char c = waitKey(1);
if (c == 27) {
break;
}
}
waitKey(0);
return;
代码运行结果如下:
我只能说速度有点感人,我有点怕啦,当然我是在i7CPU上运行的。