Research map / Adversarial robustness and security / Attacks on text and graphs
RecoverMark: Robust Watermarking for Localization and Recovery of Manipulated Faces
Haonan An, Xiaohui Ye, Guang Hua, Yihang Tao, Hangcheng Cao, Xiangyu Yu, Yuguang Fang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| embed the face as a robust watermark in the background (RecoverMark) instead of fragile watermarks | proactive manipulation localization · CelebA, Structpix2pix, regeneration attack | localization F1: 0.105 (OmniGuard) -> 0.855 | Improved |
| RecoverMark vs OmniGuard under lattice attack (unseen) | RecoverMark · CelebA, Structpix2pix, lattice attack | localization F1: 0.113 -> 0.842 | Improved |
| RecoverMark vs EditGuard under lattice attack | RecoverMark · CelebA, SD inpaint, lattice attack | localization F1: 0.704 -> 0.756 | Improved |
| RecoverMark vs OmniGuard under patch removal | RecoverMark · CelebA, SD inpaint, patch removal | localization F1: 0.911 -> 0.569 | Hurt |
| RecoverMark vs Imuge+ for face recovery under regeneration | RecoverMark · CelebA, Structpix2pix, regeneration attack | recovery PSNR (dB): 7.252 -> 22.154 | Improved |
| RecoverMark vs Imuge+ for face recovery under patch removal | RecoverMark · CelebA, Structpix2pix, patch removal | recovery PSNR and MS-SSIM: PSNR 14.445 -> 16.624; MS-SSIM 0.673 -> 0.467 | Mixed |
| train on CelebA, test on OOD FFHQ | RecoverMark · FFHQ (OOD), Instructpix2pix, regeneration attack | localization F1: 0.855 -> 0.884 | No clear effect |
| add the distortion layer (stage 2 robustness training) (ablation) | RecoverMark · CelebA, removal attacks | face recovery (qualitative) | Improved |
| introduce regeneration attack first in progressive training (ablation) | RecoverMark · CelebA, regeneration attack | robustness (qualitative) | Improved |
| increase face-to-image area ratio above 60% (ablation) | RecoverMark · capacity evaluation, regeneration attack | recovery and background PSNR | Hurt |
Closest papers by meaning
- Transferable Black-Box One-Shot Forging of Watermarks via Image Preference Models (NeurIPS 2025 · similarity 0.89)
- Image Watermarks are Removable using Controllable Regeneration from Clean Noise (ICLR 2025 · similarity 0.87)
- Invisible Image Watermarks Are Provably Removable Using Generative AI (NeurIPS 2024 · similarity 0.87)
- StableGuard: Towards Unified Copyright Protection and Tamper Localization in Latent Diffusion Models (NeurIPS 2025 · similarity 0.87)
- OmniGuard: Hybrid Manipulation Localization via Augmented Versatile Deep Image Watermarking (CVPR 2025 · similarity 0.87)
- Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
