CAM-VFD: Cross-Attention Multimodal Video Forgery Detection

In three linesCAM-VFD detects video deepfakes by analyzing cross-modal contradictions (appearance, motion, depth) using cross-attention fusion. Achieves 95.31% on GenVidBench and 93.43% on GenVideo with robustness to compression and adversarial perturbations.

Summary generated by Claude — human-verified