Zero-shot Transfer

The article presents a novel method for cross-domain face anti-spoofing (FAS) leveraging language guidance. The authors demonstrate that initializing Vision Transformer (ViT) models with multimodal pre-trained weights improves generalizability for…