A Comprehensive Study of Face Recognition Using Deep Learning

With the advent of deep learning, the performance of face recognition has been dramatically improved. On the other hand, there are few reports that discuss why the performance has been improved. In this paper, through comprehensive ex-periments, we analyze which regions are important in CNN-based face recognition. We employ the major four CNNs, AlexNet, ResNet, and EfficientNet, to be able to perform face recognition and three CNN visualization methods, Grad-CAM, Grad-CAM++, and Score-CAM, to visualize the regions in the face image that are emphasized by each face recognition method.