Research map / Image generation / Autoregressive and latent generation
Learning Robust Image Watermarking with Lossless Cover Recovery
Jiale Chen, Wei Wang, Chongyang Shi, Li Dong, Xiping Hu
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 under/overflow penalty loss with weight 10^2 (ablation) | CRMark · Auxiliary bitstream length | bits: 27828.87 -> 25807.48 | Improved |
| add under/overflow penalty loss with weight 10^4 (ablation) | CRMark · Auxiliary bitstream length | bits: 27828.87 -> 11276.45 | Improved |
| add under/overflow penalty loss with weight 10^6 (ablation) | CRMark · Auxiliary bitstream length | bits: 27828.87 -> 1962.19 | Improved |
| add latent variable regularization loss (ablation) | CRMark · lossless recovery rate | failure rate: 9.67% of stego images fail without it | Improved |
| increase penalty weight to 10^6 raises latent z bits (ablation) | CRMark · Latent variable bitstream | bits: 151.97 -> 382.61 | Hurt |
| use integer quantized flow instead of real-valued flow (R-CRMark) | CRMark · Imperceptibility | PSNR: 33.92 -> 40.95 | Improved |
| use integer invertible network instead of real-valued flow FIN | CRMark · Gaussian blur sigma=7 | extraction accuracy: 47.73 -> 93.87 | Improved |
| compare CRMark to MBRS under dropout | CRMark · Dropout p=0.7 | extraction accuracy: 96.61 -> 92.35 | Hurt |
| compare CRMark to MBRS under Gaussian noise | CRMark · Gaussian noise sigma=0.25 | extraction accuracy: 90.18 -> 88.87 | Hurt |
| compare CRMark to STDM/PZM under JPEG Qf=50 | CRMark · JPEG Qf=50 | extraction accuracy: 100.0 -> 99.27 | Hurt |
| use CRMark instead of PZM for embedding time | CRMark · Embedding time | seconds: 248.65 -> 4.51 | Improved |
| use CRMark instead of STDM for auxiliary bitstream | CRMark · Auxiliary bitstream length | bits: 56968.52 -> 1962.19 | Improved |
Closest papers by meaning
- END^2: Robust Dual-Decoder Watermarking Framework Against Non-Differentiable Distortions (AAAI 2025 · similarity 0.87)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.87)
- Image Watermarks are Removable using Controllable Regeneration from Clean Noise (ICLR 2025 · similarity 0.86)
- FreqMark: Invisible Image Watermarking via Frequency Based Optimization in Latent Space (NeurIPS 2024 · similarity 0.86)
- An Undetectable Watermark for Generative Image Models (ICLR 2025 · similarity 0.86)
- Decoder Gradient Shield: Provable and High-Fidelity Prevention of Gradient-Based Box-Free Watermark Removal (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.
