Research map / Image generation / Image restoration
Dual-level Prototype Learning for Composite Degraded Image Restoration
Zhongze Wang, Haitao Zhao, Lujian Yao, Jingchao Peng, Kaijie Zhao
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 degradation-level prototypes Pdeg to baseline (ablation) | Baseline U-Net · VCD | PSNR (dB): 20.59 -> 21.10 | Improved |
| add factor-level prototypes Pfac on top of Pdeg (ablation) | DPL · VCD | PSNR (dB): 21.10 -> 22.71 | Improved |
| add DSETB scene cross attention (ablation) | DPL · VCD | PSNR (dB): 22.71 -> 23.08 | Improved |
| add prototype scatter loss Lpsl (ablation) | DPL · VCD | PSNR (dB): 23.08 -> 25.27 | Improved |
| add within-degradation-farthest scatter constraint to loss (vs Lvar1) (ablation) | Lpsl · VCD | PSNR (dB): 19.68 -> 25.27 | Improved |
| use inverse ratio-trace form (vs Lvar2 without inverse) (ablation) | Lpsl · VCD | PSNR (dB): 21.19 -> 25.27 | Improved |
| use ratio-trace form instead of difference form (Lvar3) (ablation) | Lpsl · VCD | PSNR (dB): 23.62 -> 25.27 | Improved |
| use Lpsl instead of contrastive loss (ablation) | DPL · VCD | PSNR (dB): 24.81 -> 25.27 | Improved |
| use Lpsl instead of triplet loss (ablation) | DPL · VCD | PSNR (dB): 24.03 -> 25.27 | Improved |
| use Lpsl instead of center loss (ablation) | DPL · VCD | PSNR (dB): 23.08 -> 25.27 | Improved |
| dual-level prototype scene embeddings vs OneRestore fixed text embeddings (ablation) | OneRestore · CDD-11 | PSNR (dB): 28.47 -> 29.55 | Improved |
| replace Lpsl with contrastive loss (SSIM) (ablation) | DPL · VCD | SSIM: 0.802 -> 0.799 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- OneRestore — OneRestore: A Universal Restoration Framework for Composite Degradation (ECCV 2024) · better on 8 of 8 shared results
- PromptIR — PromptIR: Prompting for All-in-One Image Restoration (NeurIPS 2023) · better on 3 of 3 shared results
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 3 of 3 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
- OneRestore: A Universal Restoration Framework for Composite Degradation (ECCV 2024 · similarity 0.90)
- Learning Where and What to Restore for Composite Image Restoration (NeurIPS 2026 · similarity 0.88)
- Image Restoration via Multi-Domain Learning (TPAMI 2026 · similarity 0.88)
- Boosting Filter Optimization and Prompt-Guided Decoding for Mixed Degradation Image Restoration (IJCV 2026 · similarity 0.88)
- All-in-One Transformer for Image Restoration Under Adverse Weather Degradations (TPAMI 2026 · similarity 0.88)
- GenDeg: Diffusion-based Degradation Synthesis for Generalizable All-In-One Image Restoration (CVPR 2025 · similarity 0.87)
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.
