Research map / Image generation / Image restoration
Tackling Ill-Posedness of Reversible Image Conversion With Well-Posed Invertible Network
Yuanfei Huang, Hua Huang
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 L_shift to the WIC layer (ablation) | WIN-Naive · DIV2K single hiding, cover/stego | PSNR: 37.09 -> 48.62 | Improved |
| add WIC layer (without L_shift) (ablation) | INN baseline · DIV2K single hiding, cover/stego | PSNR: 37.34 -> 37.09 | Hurt |
| add WIC layer (without L_shift) (ablation) | INN baseline · DIV2K single hiding, secret/recover | PSNR: 42.65 -> 44.22 | Improved |
| add L_Det on top of WIC and L_shift (ablation) | WIN-Naive · DIV2K single hiding, cover/stego | PSNR: 48.62 -> 48.76 | Improved |
| add long-term memory (ablation) | WIN · DIV2K single hiding, cover/stego | PSNR: 48.76 -> 50.05 | Improved |
| add long-term memory (ablation) | WIN · DIV2K single hiding, secret/recover | PSNR: 49.13 -> 49.46 | Improved |
| lower lambda1 from 2 to 0.5 (ablation) | WIN · DIV2K single hiding, cover/stego | PSNR: 48.76 -> 44.72 | Hurt |
| raise lambda2 from 1 to 4 (ablation) | WIN · DIV2K single hiding, cover/stego | PSNR: 48.76 -> 45.21 | Hurt |
| raise lambda3 from 0.1 to 10 (ablation) | WIN · DIV2K single hiding, cover/stego | PSNR: 48.76 -> 48.62 | No clear effect |
| raise lambda4 from 1 to 4 (ablation) | WIN · DIV2K single hiding, cover/stego | PSNR: 48.76 -> 46.23 | Hurt |
| use WIN instead of StegFormer-B (30x fewer parameters) | 4-image hiding · DIV2K stego | PSNR: 40.18 -> 42.52 | Improved |
| use WIN instead of IRN for x2 rescaling | image rescaling · DIV2K | PSNR: 40.87 -> 41.97 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- IICNet — IICNet: A Generic Framework for Reversible Image Conversion (ICCV 2021) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- IICNet: A Generic Framework for Reversible Image Conversion (ICCV 2021 · similarity 0.91)
- Invertible generative models for inverse problems: mitigating representation error and dataset bias (ICML 2020 · similarity 0.87)
- Reversible Decoupling Network for Single Image Reflection Removal (CVPR 2025 · similarity 0.85)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.85)
- Blind Image Restoration via Fast Diffusion Inversion (NeurIPS 2024 · similarity 0.84)
- One-Step Effective Diffusion Network for Real-World Image Super-Resolution (NeurIPS 2024 · 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.
