Research map / Image generation / Image restoration
Dynamically Unfolding Recurrent Restorer: A Moving Endpoint Control Method for Image Restoration
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 moving-endpoint recurrent restorer with a learned stopping policy | blind image denoising · BSD68, σ=25–55 | PSNR: 29.16 vs 29.15; 27.72 vs 27.66; 26.71 vs 26.62; 25.91 vs 25.80 | Helped |
| use a moving-endpoint recurrent restorer with a learned stopping policy | blind image denoising · BSD68, σ=65 and σ=75 (unseen) | PSNR: 25.26 vs 23.40; 24.71 vs 18.73 | Helped |
| use a moving-endpoint recurrent restorer with a learned stopping policy | JPEG deblocking · LIVE1, QF 20 and 30 | PSNR: 31.68 vs 31.59; 33.05 vs 32.98 | Helped |
| use a moving-endpoint recurrent restorer with a learned stopping policy | JPEG deblocking · LIVE1, QF 10 (unseen) | PSNR: 29.40 vs 29.23 | Hurt |
| use a moving-endpoint recurrent restorer with a learned stopping policy | JPEG deblocking · LIVE1, QF 40 (unseen) | PSNR: 34.01 vs 33.96 | Helped |
| assign degradation-specific loop times during restoration-unit training instead of a fixed loop time (ablation) | DURR restoration unit · BSD68, σ=25, 30, 35, 40, 45, 50, 55 | peak PSNR: 28.61 -> 29.14; 28.13 -> 28.33; 27.62 -> 27.67; 27.19 -> 27.19; 26.57 -> 27.69; 26.17 -> 26.61; 24.00 -> 25.88 | Helped |
| train on multiple noise levels with a fixed loop time rather than a degradation-specific policy (ablation) | DURR restoration unit · BSD68, σ=25, 30, 35, 40, 45, 50, 55 | peak PSNR: 27.74, 27.17, 26.66, 26.24, 26.75, 25.61, 24.75 vs 29.14, 28.33, 27.67, 27.19, 27.69, 26.61, 25.88 | Hurt |
| shrink both units and apply the enhancement unit twice at each restoration stage (ablation) | Doppio-DURR · BSD68, σ=15, 25, 35, 45, 55 | PSNR: 31.32 -> 31.19; 29.28 -> 29.19; 27.84 -> 27.72; 26.83 -> 26.72; 25.80 -> 25.71 | Hurt |
| shrink both units and apply the enhancement unit twice at each restoration stage (ablation) | Doppio-DURR · BSD68, σ=15, 25, 35, 45, 55 | inference time (ms): 2.66 -> 1.28; 4.69 -> 2.31; 6.75 -> 3.01; 9.78 -> 3.79; 13.09 -> 4.65 | Helped |
| continue applying the restoration unit beyond its peak-PSNR iteration (ablation) | DURR restoration unit · BSD68 denoising | PSNR | Hurt |
Closest papers by meaning
- UP-Restorer: When Unrolling Meets Prompts for Unified Image Restoration (AAAI 2025 · similarity 0.86)
- Visual-Instructed Degradation Diffusion for All-in-One Image Restoration (CVPR 2025 · similarity 0.85)
- UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control (ICML 2025 · similarity 0.85)
- Degradation-Aware Feature Perturbation for All-in-One Image Restoration (CVPR 2025 · similarity 0.84)
- Test-Time Degradation Adaptation for Open-Set Image Restoration (ICML 2024 · similarity 0.84)
- Reversing Flow for Image Restoration (CVPR 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.