George Anjith ; Ecabert Christophe ; Shahreza Hatef Otroshi ; Kotwal Ketan ; Marcel Sebastien

Abstract

In this paper, we present EdgeFace, a lightweight and efficient facerecognition network inspired by the hybrid architecture of EdgeNeXt. Byeffectively combining the strengths of both CNN and Transformer models, and alow rank linear layer, EdgeFace achieves excellent face recognition performanceoptimized for edge devices. The proposed EdgeFace network not only maintainslow computational costs and compact storage, but also achieves high facerecognition accuracy, making it suitable for deployment on edge devices.Extensive experiments on challenging benchmark face datasets demonstrate theeffectiveness and efficiency of EdgeFace in comparison to state-of-the-artlightweight models and deep face recognition models. Our EdgeFace model with1.77M parameters achieves state of the art results on LFW (99.73%), IJB-B(92.67%), and IJB-C (94.85%), outperforming other efficient models with largercomputational complexities. The code to replicate the experiments will be madeavailable publicly.