基于Pytorch和Opencv的人脸关键点检测

Face key detection based on Pytorch and Opencv

在人脸识别技术中最重要的一步就是关键点检测,为实现简单的人脸面部关键点检测,设计一种卷积神经网络,在深度学习Pytorch框架下使用Youtube face数据集经过训练得到模型分类器,且测试集上错误率达到0.139%,结合Opencv的级联人脸检测器和训练好的Pytorch模型可定位任意图片的人脸框以及68个关键点的位置。测试结果表明,人脸及人脸关键点检测识别准确度较高,且该方法简单高效,可用于现实应用场景的模块构建。

The most important step in face recognition technology is key point detection. To realize simple face key point detection, a convolutional neural network is designed. Under the deep learning Pytorch framework, the Youtube face data set is used to train the model classification. And the error rate on the test set reaches 0.139%. Combined with Opencv’s cascaded face detector and trained Pytorch model, you can locate the face frame of any picture and the position of 68 key points. The test results show that the recognition accuracy of face and face key points is high, and the method is simple and efficient, and can be used for module construction of real application scenarios.