Research map / Image generation / Autoregressive and latent generation
Semantic Watermarking Reinvented: Enhancing Robustness and Generation Quality with Fourier Integrity
Sung Ju Lee, Nam Ik Cho
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 Hermitian symmetry (SFW) with real-only detection (ablation) | Tree-Ring · MS-COCO identification | accuracy: 0.416 -> 0.775 | Improved |
| add Hermitian symmetry (SFW) with real-only detection (ablation) | Tree-Ring · MS-COCO verification | TPR@1%FPR: 0.805 -> 0.936 | Improved |
| add imaginary component to detection once SFW preserves frequency integrity (ablation) | HSTR · MS-COCO identification | accuracy: 0.775 -> 0.889 | Improved |
| use both real and imaginary detection without SFW instead of real only (ablation) | Tree-Ring · MS-COCO identification | accuracy: 0.416 -> 0.114 | Hurt |
| add center-aware embedding on top of SFW (ablation) | HSTR · MS-COCO cropping attacks | identification accuracy: 0.374 -> 1.000 | Improved |
| add center-aware embedding on top of SFW (ablation) | HSTR · MS-COCO signal processing attacks | identification accuracy: 0.856 -> 0.838 | Hurt |
| add center-aware embedding on top of SFW (ablation) | HSTR · MS-COCO regeneration attacks | identification accuracy: 0.812 -> 0.878 | Improved |
| add SFW to Tree-Ring (no center-aware) (ablation) | Tree-Ring · MS-COCO | FID: 26.418 -> 25.071 | Improved |
| add SFW and center-aware embedding to Tree-Ring (average) (ablation) | Tree-Ring · MS-COCO identification | average accuracy: 0.114 -> 0.889 | Improved |
| add SFW to Tree-Ring latent noise | Tree-Ring · latent normality, 1,000 samples | KS failure rate: 0.234 -> 0.071 | Improved |
| use HSQR instead of RingID high-energy ring pattern | HSQR · MS-COCO generation quality | FID: 27.052 -> 24.895 | Improved |
| increase diffusion regeneration noise step from 20 to 180 (ablation) | HSTR · MS-COCO identification | accuracy: 0.999 -> 0.781 | Hurt |
Closest papers by meaning
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · similarity 0.88)
- Harnessing Frequency Spectrum Insights for Image Copyright Protection Against Diffusion Models (CVPR 2025 · similarity 0.86)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.86)
- Shallow Diffuse: Robust and Invisible Watermarking through Low-Dim Subspaces in Diffusion Models (NeurIPS 2025 · similarity 0.86)
- A Semantic Invariant Robust Watermark for Large Language Models (ICLR 2024 · similarity 0.86)
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.85)
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.
