Convolutional Neural Networks for Mobile Face Recognition with Hierarchical Feature integration
Xinxin Li, Chen Danyang, LI Yong-jie, Yan Shuming, Lin Lan, Zhang Xiangyu · 2020
Significant advances in face recognition tasks have been made with the development of convolutional neural networks. With large scale convolutional neural networks we are now able to achieve 99% accuracy on many public face test sets. However, the face recognition networks commonly used for mobile are limited by their narrow network width which makes it difficult to learn better distributed features in face recognition feature embedding learning. In this work, based on MobileFaceNets we propose an attention-guided hierarchical feature fusion model. Compared with the traditional cascading convolutional neural network classification model, our approach can explicitly consider both global and local features. And our approach requires only a small increase in computation to obtain better results than directly increasing the feature dimension.