Research map / Adversarial robustness and security / Certified and empirical robustness
Decoupling Defense Strategies for Robust Image Watermarking
Jiahui Chen, Zehang Deng, Zeyu Zhang, Chaoyang Li, Lianchen Jia, Lifeng Sun
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 |
|---|---|---|---|
| add stage 1 encoder adversarial fine-tuning (ablation) | AdvMark · COCO WEvade | bit accuracy: 0.50 -> 0.98 | Improved |
| add stage 1 encoder adversarial fine-tuning (ablation) | AdvMark · COCO | PSNR: 34.7 -> 37.0 | Improved |
| add stage 2 image optimization (ablation) | AdvMark · COCO JPEG | bit accuracy: 0.88 -> 0.99 | Improved |
| add stage 2 image optimization (ablation) | AdvMark · COCO combined distortions | bit accuracy: 0.65 -> 0.83 | Improved |
| add stage 2 image optimization (ablation) | AdvMark · COCO regeneration V1-4 | bit accuracy: 0.54 -> 0.87 | Improved |
| increase encoder fine-tuning iterations from 10 to 20 (ablation) | AdvMark · COCO combined distortions | bit accuracy: 0.83 -> 0.75 | Hurt |
| reduce encoder fine-tuning iterations from 10 to 5 (ablation) | AdvMark · COCO WEvade | bit accuracy: 0.98 -> 0.91 | Hurt |
| raise image loss weight from 5 to 7 (ablation) | AdvMark · COCO | PSNR: 37.0 -> 37.2 | Improved |
| apply AdvMark two-stage defense to HiDDeN (ablation) | HiDDeN · COCO WEvade | bit accuracy: 0.57 -> 0.91 | Improved |
| raise joint adversarial training strength S from 1.0 to 2.0 (ablation) | MBRS · COCO | PSNR: 32.1 -> 28.1 | Hurt |
| lower joint adversarial training strength S from 1.0 to 0.6 (ablation) | MBRS · COCO regeneration V1-4 | bit accuracy: 0.70 -> 0.60 | Hurt |
| add stage 2 to Stable Signature (ablation) | Stable Signature · COCO WEvade | bit accuracy: 0.78 -> 0.76 | No clear effect |
Closest papers by meaning
- END^2: Robust Dual-Decoder Watermarking Framework Against Non-Differentiable Distortions (AAAI 2025 · similarity 0.89)
- Decoder Gradient Shield: Provable and High-Fidelity Prevention of Gradient-Based Box-Free Watermark Removal (CVPR 2025 · similarity 0.89)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.89)
- ROBIN: Robust and Invisible Watermarks for Diffusion Models with Adversarial Optimization (NeurIPS 2024 · similarity 0.87)
- Leveraging Optimization for Adaptive Attacks on Image Watermarks (ICLR 2024 · similarity 0.87)
- Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks (ICLR 2024 · similarity 0.86)
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.
