RetinaFace：C++调用caffemodel的极简例程
RetinaFace的人脸检测模型，据说性价比优于MTCNN，想做成dll让其他部分调用（虽然dll还不会做）
今天实现了C++调用，不需要安装caffe，也不需要CUDA，装个opencv4.0+就可以了。
需要的有：
- Windows10
- vs2017
- opencv4.2.0
1、新建项目，配置opencv库
vs2017里面Visual C++新建项目，空项目。
我一般习惯把opencv安装在C:/opencv/里面，opencv的安装包文件名是：opencv-4.2.0-vc14_vc15.exe，安装到C:/opencv/。
解压结束以后添加系统环境变量：C:\opencv\opencv\build\x64\vc14\lib和C:\opencv\opencv\build\x64\vc14\lib和C:\opencv\opencv\build\x64\vc14\lib
给空项目配置opencv库：
a、项目使用Release x64
b、常规：平台工具集->Visual Studio 2017 (v141)
c、VC++目录->包含目录->C:\opencv\opencv\build\include\opencv2
d、VC++目录->库目录->C:\opencv\opencv\build\x64\vc14\lib
e、链接器 -> 输入 -> 附加依赖项 -> opencv_world420.lib
f、然后把这4个文件复制到main.cpp同级目录：
- opencv_world420.dll
- opencv_world420.pdb
- opencv_world420d.dll
- opencv_world420d.pdb
2、编写main.cpp
需要注意的点：
- 准备好deploy.prototxt和Widerface-RetinaFace.caffemodel
- 解析net.forward的结果费了我好大的功夫
#include
#include
#include
#include
using namespace cv;
using namespace cv::dnn;
#include
#include
#include
using namespace std;
int main()
{
CV_TRACE_FUNCTION();
String modelTxt = "deploy.prototxt";
String modelBin = "Widerface-RetinaFace.caffemodel";
String imageFile = "image_F1.jpg";
Net net = dnn::readNetFromCaffe(modelTxt, modelBin);
if (net.empty())
{
std::cerr << "Can't load network by using the following files: " << std::endl;
std::cerr << "prototxt: " << modelTxt << std::endl;
std::cerr << "caffemodel: " << modelBin << std::endl;
exit(-1);
}
Mat img = imread(imageFile, cv::IMREAD_COLOR);
int width = img.cols;
int height = img.rows;
int dims = img.channels();
if (img.empty())
{
std::cerr << "Can't read image from the file: " << imageFile << std::endl;
exit(-1);
}
//RetinaFaceNet accepts 640x640 RGB-images
Mat inputBlob = blobFromImage(img, 1, Size(640, 640), Scalar(104, 117, 123)); //Convert Mat to batch of images
Mat out;
cv::TickMeter t;
CV_TRACE_REGION("forward");
net.setInput(inputBlob, "data"); //set the network input
t.start();
// out.shape = [1,1,200,7]
out = net.forward("detection_out"); //compute output
cv::Mat detectionMat(out.size[2], out.size[3], CV_32F, out.ptr());
vector> bboxes;
//TODO：优化循环
for (int i = 0; i < detectionMat.rows; i++)
{
float confidence = detectionMat.at(i, 2);
if (confidence > 0.7)
{
int x1 = static_cast(detectionMat.at(i, 3) * width);
int y1 = static_cast(detectionMat.at(i, 4) * height);
int x2 = static_cast(detectionMat.at(i, 5) * width);
int y2 = static_cast(detectionMat.at(i, 6) * height);
vector box = { x1, y1, x2, y2 };
bboxes.push_back(box);
cv::rectangle(img, cv::Point(x1, y1), cv::Point(x2, y2), cv::Scalar(0, 255, 0), 2, 4);
//std::cout << x1 <<","<< y1<<"," << x2 <<","<< y2 << std::endl;
}
}
//cv::imwrite("x.jpg", img);
t.stop();
std::cout << "Time: " << (double)t.getTimeMilli() / t.getCounter() << " ms (average from " << t.getCounter() << " iterations)" << std::endl;
system("pause");
return 0;
}
最后生成一下解决方案就可以了，后面会生成dll了再回来补~