Research map / Image generation / Image restoration
Universal Compressed Image Restoration via Codec-Aware Conditioning with Reinforcement Learning
Changwoo Han, Hongil Kim, Donghyun Kim, Sung-Chang Lim, Seung-Won Jung
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 stage-2 RL fine-tuning | UCIR · LIVE1 (JPEG QF 5/70, D2) | PSNR: 29.29 -> 30.38 | Improved |
| add stage-2 RL fine-tuning | UCIR · LIVE1 (WebP, D2) | PSNR: 29.40 -> 30.24 | Improved |
| add stage-2 RL fine-tuning | UCIR · LIVE1 (AVC unseen, Du) | PSNR: 32.17 -> 32.60 | Improved |
| add stage-2 RL fine-tuning | UCIR · LIVE1 (JPEG->VVC double, Du) | PSNR: 27.98 -> 28.02 | No clear effect |
| add stage-2 RL fine-tuning | UCIR · LIVE1 (JPEG QF 10-40, D1) | PSNR: 30.77 -> 30.77 | No clear effect |
| add policy and value RL losses on top of reconstruction loss (ablation) | UCIR · LIVE1 (all D2 settings) | PSNR: 31.60 -> 31.75 | Improved |
| use reconstruction-loss fine-tuning on D2 without RL (ablation) | UCIR · LIVE1 (all D2 settings) | PSNR: 30.64 -> 31.60 | Improved |
| add codec type embedding to conditioning (ablation) | UCIR · LIVE1 (all D1 settings) | PSNR: 31.09 -> 31.13 | Improved |
| add compression level feature to conditioning (ablation) | UCIR · LIVE1 (all D1 settings) | PSNR: 31.09 -> 31.12 | Improved |
| combine codec type and level conditions (ablation) | UCIR · LIVE1 (all D1 settings) | PSNR: 31.07 -> 31.17 | Improved |
| compare with DFPIR on D1 JPEG | UCIR · Urban100 (JPEG QF 10-40, D1) | PSNR: 29.86 -> 30.05 | Improved |
Closest papers by meaning
- Crafting a Toolchain for Image Restoration by Deep Reinforcement Learning (CVPR 2018 · similarity 0.89)
- MoE-DiffIR: Task-customized Diffusion Priors for Universal Compressed Image Restoration (ECCV 2024 · similarity 0.86)
- Learning Image-Adaptive Codebooks for Class-Agnostic Image Restoration (ICCV 2023 · similarity 0.86)
- UP-Restorer: When Unrolling Meets Prompts for Unified Image Restoration (AAAI 2025 · similarity 0.86)
- Learning Where and What to Restore for Composite Image Restoration (NeurIPS 2026 · similarity 0.86)
- IRPO: Boosting Image Restoration via Post-training GRPO (NeurIPS 2026 · 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.
