Research map / Image generation / Image restoration
Low-Compute Watermark Removal via Dual-Domain Natural Projection
Pragati Shuddhodhan Meshram, Varun Chandrasekaran
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 frequency-domain reconstruction before semantic refinement (ablation) | DAWN · MS-COCO Tree-Ring | attack success (%): 0.0 -> 77.4 | Improved |
| add semantic refinement (SD-v2) after frequency reconstruction (ablation) | DAWN · MS-COCO Tree-Ring | attack success (%): 27.2 -> 61.4 | Improved |
| add perceptual alignment (mean-variance) to FreqRecon + SemRefine (ablation) | DAWN · MS-COCO Tree-Ring | attack success (%): 61.4 -> 77.4 | Improved |
| add perceptual alignment (mean-variance) to FreqRecon + SemRefine (ablation) | DAWN · SDP Tree-Ring | attack success (%): 65.8 -> 70.2 | Improved |
| replace mean-variance alignment with style-transfer alignment (ablation) | DAWN · MS-COCO Tree-Ring | attack success (%): 77.4 -> 71.2 | Hurt |
| replace mean-variance alignment with style-transfer alignment (ablation) | DAWN · MS-COCO Tree-Ring | PSNR (dB): 16.1 -> 18.1 | Improved |
| swap SD-v2 refiner for SDXL (ablation) | DAWN · MS-COCO Tree-Ring | attack success (%): 77.4 -> 75.2 | No clear effect |
| use frequency UNet reconstruction instead of SD-v2 img2img regeneration (ablation) | Tree-Ring removal · 20 Tree-Ring images | detector p-value (median): 5.1e-31 -> 5.9e-6 | Improved |
| use frequency UNet reconstruction instead of SD-v2 img2img regeneration (ablation) | Tree-Ring removal · 20 Tree-Ring images | LPIPS: 0.10 -> 0.53 | Hurt |
| use prompted harmonization instead of mean-variance alignment (ablation) | DAWN · Tree-Ring (appendix) | attack success (%): 77.4 -> 20 | Hurt |
| apply DAWN instead of JPEG (Q=25) distortion | Tree-Ring watermark · Tree-Ring | attack success (%): 0 -> 74 | Improved |
| apply DAWN instead of Rotation (75) distortion | Tree-Ring watermark · Tree-Ring | PSNR (dB): 9.68 -> 14.6 | Improved |
Closest papers by meaning
- Invisible Image Watermarks Are Provably Removable Using Generative AI (NeurIPS 2024 · similarity 0.88)
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · similarity 0.88)
- Image Watermarks are Removable using Controllable Regeneration from Clean Noise (ICLR 2025 · similarity 0.87)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.87)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.86)
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · 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.
