Towards Mask-Robust Face Recognition

Tao Feng, Liangpeng Xu, Hangjie Yuan, Yongfei Zhao, Mingqian Tang, Mang Wang; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 1492-1496


In this paper, we focus on the problem of mask-robust face recognition. Facial mask usually covers a major part of face, causing a significant reduction in extracting effective features. Due to such restriction, even the most advanced face recognition models are confronted with significant challenges. In light of this, this paper attempts to provide a reliable solution. Specifically, we introduce a mask-to-face image blending approach based on UV texture mapping, and a self-learning based cleaning pipeline for processing noisy training datasets. Then, considering the impacts of the long-tail distribution and hard faces samples, a loss function named Balanced Curricular Loss is introduced. Together with a bag of tricks is briefly presented. Experimental results show that the proposed solution separately achieved 84.528% @ Mask and 88.355% @ MR-ALL in InsightFace ms1m Track, which ranks 3rd when the paper submitted.

Related Material

@InProceedings{Feng_2021_ICCV, author = {Feng, Tao and Xu, Liangpeng and Yuan, Hangjie and Zhao, Yongfei and Tang, Mingqian and Wang, Mang}, title = {Towards Mask-Robust Face Recognition}, booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops}, month = {October}, year = {2021}, pages = {1492-1496} }