Explicit Cross-Modal Correlation Learning for Generalizable Deepfake Detection

The proposed method explicitly learns cross-modal correlation from a content perspective to enhance the generalizability of deepfake detection across diverse cross-modal forgery techniques.