Differential Reconstruction-driven Residual Guidance for Face Forgery Detection: Code Implementation

First, the contrastive differential reconstruction network (CDRNet) enhances the model's ability to distinguish between authentic and forged faces by amplifying differences during the reconstruction process.

Second, the residual dual-domain guidance module (RDDGM) leverages residual reconstruction images to guide the fusion of dual-domain features, thereby improving the model's sensitivity to subtle forgery traces. Finally, the text-aware loss module (TALM) incorporates textual modality information to effectively learn universal forgery features across different forgery methods, thus enhancing the network's generalization capability toward unseen forged faces.