Balanced Masked and Standard Face Recognition

TL;DR

This paper introduces improved face recognition methods that balance performance on masked and unmasked faces by novel network design, data strategies, and training techniques, achieving robust results in masked face recognition challenges.

Contribution

It proposes new network components, data augmentation, and training strategies to balance masked and standard face recognition performance.

Findings

Achieved balanced face recognition performance on masked and unmasked faces.

Controlled masked face data to prevent overfitting.

Implemented new network modules and training strategies.

Abstract

We present the improved network architecture, data augmentation, and training strategies for the Webface track and Insightface/Glint360K track of the masked face recognition challenge of ICCV2021. One of the key goals is to have a balanced performance of masked and standard face recognition. In order to prevent the overfitting for the masked face recognition, we control the total number of masked faces by not more than 10\% of the total face recognition in the training dataset. We propose a few key changes to the face recognition network including a new stem unit, drop block, face detection and alignment using YOLO5Face, feature concatenation, a cycle cosine learning rate, etc. With this strategy, we achieve good and balanced performance for both masked and standard face recognition.