Research map / Image generation / Image restoration
Restore Anything with Masks: Leveraging Mask Image Modeling for Blind All-in-One Image Restoration
Chujie Qin, Ruiqi Wu, Zikun Liu, Xin Lin, Chun-Le Guo, Hyun Hee Park, Chongyi Li
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 MIM pre-training with 50% pixel masking and MAC fine-tuning to PromptIR | PromptIR · All-in-one, average of 7 tasks | PSNR: 28.17 -> 28.76 | Improved |
| add RAM pipeline to SwinIR | SwinIR · All-in-one, average of 7 tasks | PSNR: 26.64 -> 27.28 | Improved |
| add RAM to PromptIR on kernel deblurring | PromptIR · LSDIR-Blur | PSNR: 31.01 -> 29.57 | Hurt |
| add RAM to PromptIR on dehazing and low-light enhancement | PromptIR · SOTS and LOL | PSNR: SOTS 28.70 -> 29.64; LOL 21.19 -> 24.46 | Improved |
| increase masking ratio to 80% (ablation) | RAM-SwinIR · All-in-one, average of 7 tasks | PSNR: 27.28 -> 27.08 | Hurt |
| vary masking ratio between 20% and 60% (ablation) | RAM-SwinIR · All-in-one, average of 7 tasks | PSNR: 27.28 (20%), 27.21 (40%), 27.28 (50%), 27.26 (60%) | No clear effect |
| pre-train MIM with paired degraded/clean data instead of ground truth only (ablation) | RAM-SwinIR · All-in-one, average of 7 tasks | PSNR: 26.62 -> 27.28 | Improved |
| select fine-tuning layers by MAC instead of IG or random (ablation) | RAM-SwinIR · All-in-one, average of 7 tasks | PSNR: 26.92 (IG) -> 27.28 | Improved |
| fine-tune 100% of layers instead of 10% (ablation) | RAM-SwinIR · In-distribution average | PSNR: 27.28 -> 27.54 | Improved |
| fine-tune 100% of layers instead of 10% (ablation) | RAM-SwinIR · Out-of-distribution denoising average | PSNR: 17.99 -> 15.93 | Hurt |
| use 1x1 mask patches instead of 4x4 or 8x8 (ablation) | RAM-SwinIR pre-training · qualitative reconstruction | reconstruction quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PromptIR — PromptIR: Prompting for All-in-One Image Restoration (NeurIPS 2023) · better on 5 of 6 shared results
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 1 of 1 shared result
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
- Rethinking Expressivity and Degradation-Awareness in Attention for All-in-One Blind Image Restoration (ICLR 2026 · similarity 0.88)
- ClearAIR: A Human-Visual-Perception-Inspired All-in-One Image Restoration (AAAI 2026 · similarity 0.88)
- RestoreFormer++: Towards Real-World Blind Face Restoration From Undegraded Key-Value Pairs (TPAMI 2023 · similarity 0.87)
- Context-Aware Pretraining for Efficient Blind Image Decomposition (CVPR 2023 · similarity 0.87)
- A Survey on All-in-One Image Restoration: Taxonomy, Evaluation and Future Trends (TPAMI 2025 · similarity 0.87)
- Debiased All-in-one Image Restoration with Task Uncertainty Regularization (AAAI 2025 · 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.
