Masked LFW (MLFW) Database (whdeng.cn)

The newly launched Masked LFW (MLFW) database serves as an essential resource for evaluating face recognition systems under the challenges posed by mask-wearing, a trend significantly accelerated by the COVID-19 pandemic. Building upon the Cross-Age LFW (CALFW) database, MLFW incorporates a robust tool that generates masked faces from unmasked originals, creating variations that reflect common styles found in everyday life. This innovation addresses the notable decline in recognition accuracy seen in state-of-the-art (SOTA) models, which plummets by 5%-16% when tested on masked images compared to unmasked counterparts. The significance of MLFW lies in its structured approach to assessing masked face verification, presenting a more challenging benchmark that enables accurate evaluations of various face verification methods. By maintaining consistent data size and protocols from CALFW while introducing diverse face pair combinations, the MLFW aims to bridge the gap between reported performance on standard benchmarks and actual effectiveness in real-world scenarios. The dataset not only highlights the considerable intra-class and inter-class variance influenced by masking but also sets the stage for future advancements in the development of more reliable face recognition algorithms, ensuring that technology adapts to the evolving social landscape.