NAS-FAS: Static-Dynamic Central Difference Network Search for Face Anti-Spoofing

Face anti-spoofing (FAS) plays a vital role in securing face recognition systems. Existing methods heavily rely on the expert-designed networks, which may lead to sub-optimal solution for FAS task. Here we propose first method based neural architecture search (NAS), called NAS-FAS, discover well-suited task-aware networks. Unlike previous NAS works mainly focus developing efficient strategies generic object classification, pay more attention study spaces The challenges of utilizing are two folds: networks searched 1) specific acquisition condition might perform poorly unseen conditions, and 2) particular spoofing attacks generalize badly attacks. To overcome these issues, develop novel space consisting central difference convolution pooling operators. Moreover, an static-dynamic representation is exploited fully mining FAS-aware spatio-temporal discrepancy. Besides, Domain/Type-aware Meta-NAS, leverages cross-domain/type knowledge robust searching. Finally, order evaluate transferability cross datasets unknown attack types, release large-scale 3D mask dataset, namely CASIA-SURF 3DMask, supporting new 'cross-dataset cross-type' testing protocol. Experiments demonstrate that proposed NAS-FAS achieves state-of-the-art performance nine benchmark with four protocols.