Research map / Image generation / Super-resolution
Bridging Degradation Discrimination and Generation for Universal Image Restoration
JiaKui Hu, Zhengjian Yao, Lujia Jin, Yanye Lu
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 the bridging stage before restoration fine-tuning (ablation) | BDG · All-in-one 5D | average PSNR: 31.03 -> 32.09 | Improved |
| add the restoration fine-tuning stage after bridging (ablation) | BDG · All-in-one 5D | average PSNR: 30.25 -> 32.09 | Improved |
| add bridging stage to restoration fine-tuning (ablation) | BDG · Real-world SR | PSNR / CLIPIQA: PSNR 28.73 -> 28.80, CLIPIQA 0.4787 -> 0.6053 | Mixed |
| add degradation classification loss L_deg-cls to bridging (ablation) | BDG · All-in-one 5D | PSNR: 20.88 -> 32.09 | Improved |
| add L_bridge and L_deg-cls to L_gen (ablation) | BDG · All-in-one 5D | PSNR: 31.11 -> 32.09 | Improved |
| add L_bridge without L_deg-cls (ablation) | BDG · All-in-one 5D | PSNR: 31.11 -> 20.88 | Hurt |
| add L_bridge in RFT stage (ablation) | BDG · Real-world SR | PSNR: 28.23 -> 28.80 | Improved |
| add L_fcnl in RFT stage (ablation) | BDG · Real-world SR | PSNR: 27.88 -> 28.80 | Improved |
| use full set of angles and scales in MAS-GLCM instead of reduced angles (ablation) | BDG · All-in-one 5D | average PSNR: 31.77 -> 32.09 | Improved |
| use full angles and scales instead of sparse horizontal/vertical set (ablation) | BDG · All-in-one 5D | average PSNR: 31.93 -> 32.09 | Improved |
| use MAS-GLCM instead of Laplace gradient for degradation level classification | KNN classifier · Degradation classification | level accuracy: 20.83% -> 74.17% | Improved |
| use BDG instead of DiffUIR-L (same architecture) | BDG · All-in-one deraining | PSNR: 31.03 -> 34.75 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- SeeSR — SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024) · better on 2 of 2 shared results
- ResShift — ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS 2023) · better on 1 of 2 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
- Towards a Universal Image Degradation Model via Content-Degradation Disentanglement (ICCV 2025 · similarity 0.90)
- Diffusion Once and Done: Degradation-Aware LoRA for All-in-One Image Restoration (AAAI 2026 · similarity 0.89)
- GenDeg: Diffusion-based Degradation Synthesis for Generalizable All-In-One Image Restoration (CVPR 2025 · similarity 0.89)
- Visual-Instructed Degradation Diffusion for All-in-One Image Restoration (CVPR 2025 · similarity 0.88)
- Learning Where and What to Restore for Composite Image Restoration (NeurIPS 2026 · similarity 0.88)
- Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (AAAI 2025 · similarity 0.88)
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.
