Research map / Adversarial robustness and security / Attacks on text and graphs
PECCVAI: Overcoming the Brittleness of AI Image Watermarking Under Visual Paraphrasing Attacks
Shreyas Dixit, Ashhar Aziz, Shashwat Bajpai, Vasu Sharma, Aman Chadha, Vinija Jain, Amitava Das
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 |
|---|---|---|---|
| use XRAI saliency instead of Vanilla IG for NMP selection (ablation) | PECCAVI · paraphrase s=0.1 | avg WDP: 0.73 -> 0.90 (Top 50) | Improved |
| use XRAI saliency instead of MSI Net (ablation) | PECCAVI · paraphrase s=0.2 | avg WDP: 0.83 -> 0.85 (Top 50) | Improved |
| use XRAI saliency instead of Vanilla IG (ablation) | PECCAVI · MS COCO watermark distortion | PSNR (dB): 31.31 -> 29.84 (Top 50) | Hurt |
| increase number of saliency maps from 2 to 5 (ablation) | PECCAVI Top-30 · paraphrase s=0.1 | avg WDP: 0.877 -> 0.905 | Improved |
| increase number of saliency maps from 5 to 7 (ablation) | PECCAVI Top-30 · paraphrase s=0.1 | avg WDP: 0.905 -> 0.906 | No clear effect |
| select NMP patches by top saliency instead of random (ablation) | PECCAVI · paraphrase s=0.1 | avg WDP: 0.842 -> 0.906 (N=7) | Improved |
| PECCAVI (XRAI) vs WAM under visual paraphrase | PECCAVI · paraphrase s=0.1 | avg WDP: 0.63 -> 0.92 | Improved |
| PECCAVI (XRAI) vs ZoDiac under visual paraphrase | PECCAVI · paraphrase s=0.2 | avg WDP: 0.70 -> 0.87 | Improved |
| PECCAVI vs WAM/Stable Signature on image quality | PECCAVI · MS COCO | PSNR (dB): 46.05 -> 29.56 | Hurt |
Closest papers by meaning
- Invisible Image Watermarks Are Provably Removable Using Generative AI (NeurIPS 2024 · similarity 0.89)
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · similarity 0.88)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.88)
- Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks (ICLR 2024 · similarity 0.87)
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · similarity 0.87)
- WMCopier: Forging Invisible Watermarks on Arbitrary Images (NeurIPS 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.
