This repo is the official implementation of “Attack-Resistant Watermarking for AIGC Image Forensics via Diffusion-based Semantic Deflection”.
The complete appendix is available here. It includes:
- Full theoretical proof for the security analysis
- Complete configurations for all implemented attacks
- Ablation study on the number of deflection steps
- User study results on watermark imperceptibility and image quality
- Visual results for tampering localization under three attack types (Photoshop, deepfake, AIGC)
- Additional visualizations of watermarked images
Please refer to the appendix for in-depth experimental details and extended results.