Research map / Image generation / Autoregressive and latent generation
OptMark: Robust Multi-bit Diffusion Watermarking via Inference Time Optimization
Jiazheng Xing, Hai Ci, Hongbin Xu, Hangjie Yuan, Yong Liu, Mike Zheng Shou
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 structure watermark to detail-only watermark (ablation) | OptMark · regeneration attacks | TPR: 0.407 -> 0.872 | Improved |
| add detail watermark to structure-only watermark (ablation) | OptMark · geometric, valuemetric and editing attacks | TPR: 0.935 -> 1.000 | Improved |
| add detail watermark to structure-only watermark (ablation) | OptMark · regeneration attacks | TPR: 0.567 -> 0.872 | Improved |
| add initial constraint (normalization and Linit) (ablation) | OptMark · image quality | delta FID: 9.779 -> 7.928 | Improved |
| add low-order statistics loss Llow (ablation) | OptMark · image quality | delta FID: 7.928 -> 4.434 | Improved |
| add high-order statistics loss Lhigh (ablation) | OptMark · image quality | delta FID: 4.434 -> 3.069 | Improved |
| add quality constraints (ablation) | OptMark · robustness (avg of four attack types and none) | bit accuracy: 0.985 -> 0.983 | No clear effect |
| use OptMark instead of Gaussian Shading on regeneration attacks | OptMark · regeneration attacks | bit accuracy: 0.986 -> 0.923 | Hurt |
| use OptMark instead of SSL Watermark across all attacks | OptMark · average over attacks | TPR: 0.763 -> 0.972 | Improved |
| use OptMark watermark instead of no watermark | Stable Diffusion v2.1 · MS-COCO-2017 | FID: 124.309 -> 127.378 | Hurt |
Closest papers by meaning
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.91)
- ROBIN: Robust and Invisible Watermarks for Diffusion Models with Adversarial Optimization (NeurIPS 2024 · similarity 0.90)
- Shallow Diffuse: Robust and Invisible Watermarking through Low-Dim Subspaces in Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Secure Seed-Based Multi-bit Watermarking for Diffusion Models from First Principles (NeurIPS 2026 · similarity 0.89)
- An Undetectable Watermark for Generative Image Models (ICLR 2025 · similarity 0.89)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.88)
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.
