Research map / Architectures and efficiency / RNNs and normalization
Flow-Based Robust Watermarking with Invertible Noise Layer for Black-Box Distortions
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a shared-parameter invertible encoder/decoder instead of a separate encoder and decoder | Flow-based watermarking · JPEG compression | ACC (%): 94.9 -> 99.71 | Helped |
| Use the flow-based watermarking framework | Robust watermarking · Gaussian noise, σ=0.05 | ACC (%): 94.2 -> 98.8 | Helped |
| Use the proposed framework for black-box distortion robustness | Robust watermarking · Reverse Color, +++ | ACC (%): 65.8 -> 99.9 | Helped |
| Combine Gaussian-noise training with INL rather than Gaussian noise alone (ablation) | Proposed watermarking framework · Heavy Color | ACC (%): 94.4 -> 96.8 | Helped |
| Add Gaussian-noise pretraining to INL-only training (ablation) | Proposed watermarking framework · Heavy Color | ACC (%): 67.5 -> 96.8 | Helped |
| Combine Gaussian-noise training with INL rather than Gaussian noise alone (ablation) | Proposed watermarking framework · Sketch | ACC (%): 92.5 -> 95.1 | Helped |
| Add Gaussian-noise pretraining to INL-only training (ablation) | Proposed watermarking framework · Sketch | ACC (%): 94.9 -> 95.1 | No clear effect |
| Replace the CNN ResNet noise layer with the invertible noise layer (ablation) | Watermarking noise simulation · Heavy Color | ACC (%): 90.8 -> 95.6 | Helped |
| Replace the CNN ResNet noise layer with the invertible noise layer (ablation) | Watermarking noise simulation · Sketch | ACC (%): 79.3 -> 91.8 | Helped |
| Apply the backward pass of INL as denoising before extraction (ablation) | Proposed watermarking framework · Heavy Color | ACC (%): 95.6 -> 96.8 | Helped |
| Apply the backward pass of INL as denoising before extraction (ablation) | Proposed watermarking framework · Sketch | ACC (%): 91.8 -> 95.1 | Helped |
| Use a separately trained CNN ResNet backward denoiser (ablation) | CNN-based noise simulation · Heavy Color | ACC (%): 90.8 -> 82.5 | Hurt |
Closest papers by meaning
- END^2: Robust Dual-Decoder Watermarking Framework Against Non-Differentiable Distortions (AAAI 2025 · similarity 0.87)
- Robust Message Embedding via Attention Flow-Based Steganography (CVPR 2025 · similarity 0.86)
- IWRN:A Robust Blind Watermarking Method for Artwork Image Copyright Protection Against Noise Attack (AAAI 2025 · similarity 0.85)
- Hidden in the Noise: Two-Stage Robust Watermarking for Images (ICLR 2025 · similarity 0.84)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.84)
- T2SMark: Balancing Robustness and Diversity in Noise-as-Watermark for Diffusion Models (NeurIPS 2025 · similarity 0.84)
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.