Addressing modality unreliability and imbalance is crucial for enhancing the generalizability of multi-modal face anti-spoofing methods.
Utilizing textual prompts for face anti-spoofing improves generalization.