Research map / Image generation / Image restoration
Sim-to-Real: An Unsupervised Noise Layer for Screen-Camera Watermarking Robustness
Yufeng Wu, Xin Liao, Baowei Wang, Han Fang, Xiaoshuai Wu, Mingyue Chen, Guiling Wang
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 S2R unsupervised noise layer instead of SSDS | MCFN watermarking · S+L, 30 cm, 0 deg | BER (%): 5.1 -> 2.1 | Improved |
| use S2R noise layer instead of PIMoG | MCFN watermarking · S+L, 30 cm, 40 deg | BER (%): 9.5 -> 6.0 | Improved |
| use S2R noise layer instead of StegaStamp | MCFN watermarking · PSNR | PSNR (dB): 39.89 -> 42.27 | Improved |
| replace unsupervised S2R with supervised variant trained on 900 paired samples (ablation) | S2R · S+L, 30 cm, 0 deg | BER (%): 1.6 -> 3.8 | Hurt |
| replace the MIMO-UNet unsupervised translation with CycleGAN (ablation) | S2R · S+L, 30 cm, 0 deg | BER (%): 2.1 -> 2.9 | Hurt |
| replace the MIMO-UNet unsupervised translation with DualGAN (ablation) | S2R · S+L, 30 cm, 0 deg | BER (%): 2.1 -> 3.5 | Hurt |
| add unsupervised refinement network on top of SSDS simulation (ablation) | SSDS · S+L, 30 cm, 0 deg | BER (%): 5.0 vs 5.1 in Table 2 | Mixed |
| train S2R on a single device pair (I+E) vs combined SIM+LEA (ablation) | S2R · S+L, 30 cm, 0 deg | BER (%): 2.1 -> 1.6 | Improved |
| test S2R on device pairs not used in S2R training (ablation) | S2R · Cross-device | BER (%) | No clear effect |
| train S2R and watermark network on different datasets (ablation) | S2R · Cross-source | BER (%) | No clear effect |
| increase shooting angle to 60 degrees | S2R · S+L | BER (%): 5.8 -> 10.2 | Hurt |
Closest papers by meaning
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.87)
- Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances (ICLR 2025 · similarity 0.86)
- RoPaSS: Robust Watermarking for Partial Screen-Shooting Scenarios (AAAI 2025 · similarity 0.85)
- StableGuard: Towards Unified Copyright Protection and Tamper Localization in Latent Diffusion Models (NeurIPS 2025 · similarity 0.85)
- Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks (ICLR 2024 · 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.
