Face Transformer for Recognition

TL;DR

This paper explores the use of Transformer models for face recognition, modifying the patch generation process to improve inter-patch information, and demonstrates comparable performance to CNNs on standard benchmarks.

Contribution

It introduces a novel Face Transformer model with overlapping patches for face recognition and evaluates its performance against CNNs on multiple datasets.

Findings

Face Transformer achieves comparable accuracy to CNNs.

Modified patch generation improves inter-patch information.

Models trained on large datasets perform well on benchmarks.

Abstract

Recently there has been a growing interest in Transformer not only in NLP but also in computer vision. We wonder if transformer can be used in face recognition and whether it is better than CNNs. Therefore, we investigate the performance of Transformer models in face recognition. Considering the original Transformer may neglect the inter-patch information, we modify the patch generation process and make the tokens with sliding patches which overlaps with each others. The models are trained on CASIA-WebFace and MS-Celeb-1M databases, and evaluated on several mainstream benchmarks, including LFW, SLLFW, CALFW, CPLFW, TALFW, CFP-FP, AGEDB and IJB-C databases. We demonstrate that Face Transformer models trained on a large-scale database, MS-Celeb-1M, achieve comparable performance as CNN with similar number of parameters and MACs. To facilitate further researches, Face Transformer models and codes are available at https://github.com/zhongyy/Face-Transformer.