opencv+caffe 基于CNN的年龄性别检测实例demo（c++）

环境配置

---

visual stdio2017配置openCV
CNN使用caffe的age_net.caffemodel框架

原理讲解

---

基于2015 CVPR 的论文，《Age and Gender Classification using Convolutional Neural Networks》
参考博客：https://blog.csdn.net/qq_14845119/article/details/52454539

一些思考：

当时是作为一个小的demo展示。
关于论文老师问了我一些问题：

• 这篇文章里为什么用的是八分类，不是五分类，十分类？
• 该论文是在audience上数据集上测试，如今已经解决了很多包括光照和遮挡等限制问题，那么如何继续提升精度？我的回答是 增大数据集or对图片进行预处理，老师说说的太宽泛了，细致的我也答不上来emmmmm

代码详解

---

#include 
#include 
#include 

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

//引入模型文件(模型文件+描述文件)
//引用haar模型检测人脸
String haar_file = "G:/opencv/build/etc/haarcascades/haarcascade_frontalface_alt_tree.xml";
String age_model = "G:/VCprojects/DNN/age_gender/age_net.caffemodel";
String age_text = "G:/VCprojects/DNN/age_gender/deploy_age.prototxt";

String gender_model = "G:/VCprojects/DNN/age_gender/gender_net.caffemodel";
String gender_text = "G:/VCprojects/DNN/age_gender/deploy_gender.prototxt";

//定义方法
void predict_age(Net &net, Mat &image);
void predict_gender(Net &net, Mat &image);

int main(int argc, char** argv) {
     

	//使用.prototxt和.caffemodel文件的路径读取并初始化网络
	Net age_net = readNetFromCaffe(age_text, age_model);
	Net gender_net = readNetFromCaffe(gender_text, gender_model);

	//检查网络是否已成功读取
	if (age_net.empty())
	{
     
		std::cerr << "Can't load age network by using the following files: " << std::endl;
		std::cerr << "prototxt:   " << age_text << std::endl;
		std::cerr << "caffemodel: " << age_model << std::endl;
		std::cerr << "age_net.caffemodel can be downloaded here:" << std::endl;
		exit(-1);
	}
	if (gender_net.empty())
	{
     
		std::cerr << "Can't load gender network by using the following files: " << std::endl;
		std::cerr << "prototxt:   " << gender_text << std::endl;
		std::cerr << "caffemodel: " << gender_model << std::endl;
		std::cerr << "gender_net.caffemodel can be downloaded here:" << std::endl;
		exit(-1);
	}

	Mat src = imread("G:/VCprojects/DNN/age_gender/face00.png");
	//Mat src = imread("G:/VCprojects/DNN/age_gender/three.png");
	//Mat src = imread("G:/VCprojects/DNN/age_gender/9.jpg");
	if (src.empty()) {
     
		printf("could not load image...\n");
		return -1;
	}
	namedWindow("input", CV_WINDOW_AUTOSIZE);
	imshow("input", src);
	//创建一个级联检测器
	CascadeClassifier detector;
	//load文件
	detector.load(haar_file);
	//vector存放人脸检测的结果
	vector<Rect> faces;
	Mat gray;
	//变成一张灰度图像
	cvtColor(src, gray, COLOR_BGR2GRAY);
	//检测人脸
	detector.detectMultiScale(gray, faces, 1.02, 1, 0, Size(40, 40), Size(200, 200));

	//找到人脸的话
	for (size_t t = 0; t < faces.size(); t++) {
     
		//找到的人脸在他上面绘制一个矩形
		rectangle(src, faces[t], Scalar(0, 255, 255), 2, 8, 0);
		//预测年龄和性别
		predict_age(age_net, src(faces[t]));
		predict_gender(age_net, src(faces[t]));
	}
	imshow("age-gender-prediction-demo", src);

	waitKey(0);
	return 0;
}

//年龄的8个分类
vector<String> ageLabels() {
     
	vector<String> ages;
	ages.push_back("0-2");
	ages.push_back("4 - 6");
	ages.push_back("8 - 13");
	ages.push_back("15 - 20");
	ages.push_back("25 - 32");
	ages.push_back("38 - 43");
	ages.push_back("48 - 53");
	ages.push_back("60-");
	return ages;
}


//预测年龄
void predict_age(Net &net, Mat &image) {
     
	//输入
	//读取输入图像并转换为Blob，可由Net接受
	Mat blob = blobFromImage(image, 1.0, Size(227, 227));
	//将blob传递到网络
	net.setInput(blob, "data");
	//Make forward pass 向前传播
	Mat prob = net.forward("prob");
	//一维一行，做8个年龄分类
	Mat probMat = prob.reshape(1, 1);
	//求max_location
	Point classNum;
	double classProb;

	vector<String> ages = ageLabels();
	//求最大可能性
	minMaxLoc(probMat, NULL, &classProb, NULL, &classNum);
	//获取年龄
	int classidx = classNum.x;
	//打印结果
	putText(image, format("age:%s", ages.at(classidx).c_str()), Point(2, 10), FONT_HERSHEY_PLAIN, 0.8, Scalar(0, 0, 255), 1);
	std::cout << "Best class at age: #" << classidx << " '" << ages.at(classidx) << "'" << std::endl;
	std::cout << "Probability: " << classProb * 100 << "%" << std::endl;
}

void predict_gender(Net &net, Mat &image) {
     
	// 输入
	Mat blob = blobFromImage(image, 1.0, Size(227, 227));
	net.setInput(blob, "data");
	//预测分类
	Mat prob = net.forward("prob");
	Mat probMat = prob.reshape(1, 1);
	//保证要么是male 要么是female，promat。at（0,0）或者（0,1）
	putText(image, format("gender:%s", (probMat.at<float>(0, 0) > probMat.at<float>(0, 1) ? "M" : "F")),
		Point(2, 20), FONT_HERSHEY_PLAIN, 0.8, Scalar(0, 0, 255), 1);
}

测试输出