Research map / Reinforcement learning / Visual and representation RL
Crafting a Toolchain for Image Restoration by Deep Reinforcement Learning
Ke Yu, Chao Dong, Liang Lin, Chen Change Loy
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 6 rather than 12 restoration tools (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 25.72 | Hurt |
| increase toolbox size from 12 to 18 tools (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 26.17 | No clear effect |
| limit toolchain length to 2 rather than 3 (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 25.99 | Hurt |
| allow toolchain length 4 rather than 3 (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 26.20 | No clear effect |
| add slight noise and compression to tool training data (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 25.91 -> 26.20 | Improved |
| jointly fine-tune tools using agent-selected toolchains (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 26.45 | Improved |
| replace stepwise PSNR reward with stepwise SSIM reward (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 25.20 | Hurt |
| replace stepwise PSNR reward with final PSNR reward (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 26.11 | No clear effect |
| replace stepwise PSNR reward with final MSE reward (ablation) | RL-Restore · DIV2K test Moderate | SSIM: 0.5441 -> 0.5166 | Hurt |
| remove the agent's stopping action (ablation) | RL-Restore · DIV2K test Moderate | PSNR: 26.20 -> 26.08 | Hurt |
| use the full RL-Restore toolchain instead of VDSR-s | DIV2K test Severe (unseen) · DIV2K test Severe (unseen) | PSNR: 24.50 -> 25.20 | Improved |
Closest papers by meaning
- TIR-Agent: Training an Explorative and Efficient Agent for Image Restoration (ECCV 2026 · similarity 0.90)
- Path-Restore: Learning Network Path Selection for Image Restoration (TPAMI 2021 · similarity 0.89)
- OPERA: An Agent for Image Restoration with End-to-End Joint Planning–Execution Optimization (NeurIPS 2026 · similarity 0.87)
- IRPO: Boosting Image Restoration via Post-training GRPO (NeurIPS 2026 · similarity 0.86)
- Residual-Conditioned Optimal Transport: Towards Structure-Preserving Unpaired and Paired Image Restoration (ICML 2024 · similarity 0.85)
- Controlling Vision-Language Models for Multi-Task Image Restoration (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.
