Two-Step Data Augmentation for Masked Face Detection and Recognition: Turning Fake Masks to Real
The research proposes a two-step data augmentation pipeline to improve masked face recognition by generating realistic synthetic masks. It first uses rule-based warping to create initial masks, then employs an adapted AttentionGAN model to translate these into highly realistic outputs. This method significantly enhances mask realism compared to simple warping and other existing GAN methods.