Research map / Adversarial robustness and security / Attacks on text and graphs
ARIW-Framework: Adaptive Robust Iterative Watermarking Framework
Shaowu Wu, Liting Zeng, Wei Lu
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 ARIW instead of prior watermarking methods (alpha=1.0) | deep watermarking · Mirflickr (D1) | average bit accuracy (%): 99.78 -> 99.90 (vs StegaStamp) | Improved |
| compare PSNR to MBRS | MBRS · Mirflickr (D1) | PSNR (dB): 41.99 -> 42.084 | Improved |
| increase embedding strength alpha from 0.2 to 2.0 (ablation) | ARIW · Mirflickr (D1) | PSNR (dB): 50.48 -> 36.71 | Hurt |
| increase embedding strength alpha from 0.2 to 1.0 (ablation) | ARIW · Mirflickr (D1) | JPEG bit accuracy (%): 75.78 -> 99.66 | Improved |
| use image-gradient adaptive embedding strength (ablation) | ARIW · Mirflickr (D1) | Crop bit accuracy (%): 91.13 -> 99.71 | Improved |
| use image-gradient adaptive embedding strength (ablation) | ARIW · Mirflickr (D1) | JPEG bit accuracy (%): 99.88 -> 99.66 | No clear effect |
| initialize iteration at 1 rather than 0 (ablation) | ARIW · Mirflickr (D1) | Crop bit accuracy (%): 86.21 -> 99.71 | Improved |
| vary initial iteration state (original image or noise) (ablation) | ARIW · Mirflickr (D1) | bit accuracy: 99.71 vs 98.02 vs 99.59 (Crop) | No clear effect |
| use 3x3 kernel instead of 7x7 (ablation) | ARIW · Mirflickr (D1) | Crop bit accuracy (%): 89.27 -> 99.71 | Improved |
| use 3x3 kernel instead of 1x1 (ablation) | ARIW · Mirflickr (D1) | Crop bit accuracy (%): 95.38 -> 99.71 | Improved |
| add CBAM attention to residual (ablation) | ARIW · Mirflickr (D1) | JPEG bit accuracy (%): 99.66 -> 77.88 | Hurt |
| add spatial attention to residual (ablation) | ARIW · Mirflickr (D1) | JPEG bit accuracy (%): 99.66 -> 93.33 | Hurt |
Closest papers by meaning
- Leveraging Optimization for Adaptive Attacks on Image Watermarks (ICLR 2024 · similarity 0.86)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.86)
- Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances (ICLR 2025 · similarity 0.86)
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.84)
- CoSDA: Enhancing the Robustness of Inversion-based Generative Image Watermarking Framework (AAAI 2025 · similarity 0.84)
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.
