VarGFaceNet: An Efficient Variable Group Convolutional Neural Network for Lightweight Face Recognition

To improve the discriminative and generalization ability of lightweightnetwork for face recognition, we propose an efficient variable groupconvolutional network called VarGFaceNet. Variable group convolution isintroduced by VarGNet to solve the conflict between small computational costand the unbalance of computational intensity inside a block. We employ variablegroup convolution to design our network which can support large scale faceidentification while reduce computational cost and parameters. Specifically, weuse a head setting to reserve essential information at the start of the networkand propose a particular embedding setting to reduce parameters offully-connected layer for embedding. To enhance interpretation ability, weemploy an equivalence of angular distillation loss to guide our lightweightnetwork and we apply recursive knowledge distillation to relieve thediscrepancy between the teacher model and the student model. The champion ofdeepglint-light track of LFR (2019) challenge demonstrates the effectiveness ofour model and approach. Implementation of VarGFaceNet will be released athttps://github.com/zma-c-137/VarGFaceNet soon.