Research map / Image generation / Image restoration
Boosting Filter Optimization and Prompt-Guided Decoding for Mixed Degradation Image Restoration
Yubin Gu, Yuan Meng, Xiaoshuai Sun, Jiayi Ji, Rongrong Ji
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 LDO module to the baseline (ablation) | MDIR baseline · CIR average | PSNR: 31.030 -> 32.937 | Improved |
| add CFE module on top of baseline + LDO (ablation) | MDIR · CIR average | PSNR: 32.937 -> 33.213 | Improved |
| add CFE to IRNeXt (ablation) | IRNeXt · CIR average | PSNR: 32.568 -> 32.823 | Improved |
| add CFE to Restormer (ablation) | Restormer · CIR average | PSNR: 31.621 -> 31.813 | Improved |
| unified training over all degradations vs one-to-one training on noise (ablation) | MDIR · CIR noise | PSNR: 35.499 -> 35.736 when trained separately | Hurt |
| unified training vs one-to-one training on haze (ablation) | MDIR · CIR haze | PSNR: 30.056 -> 31.687 | Improved |
| unified training vs one-to-one training on rain+haze+noise (ablation) | MDIR · CIR rain+haze+noise | PSNR: 27.619 -> 28.458 | Improved |
| unified training vs one-to-one training on snow (ablation) | MDIR · CIR snow | PSNR: 34.811 -> 35.583 when trained separately | Hurt |
| increase LDO kernel size from 3 to 7 (ablation) | MDIR · Rain100H | PSNR: 33.086 -> 33.664 | Improved |
| increase LDO kernel size from 3 to 7 (ablation) | MDIR · Rain100L | PSNR: 39.090 -> 38.933 | Hurt |
| MDIR vs PromptIR on CIR | MDIR · CIR average | PSNR: 32.133 -> 33.213 | Improved |
| MDIR vs second-best on deraining | MDIR · Rain100H | PSNR: 31.770 -> 33.086 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- IRNeXt — IRNeXt: Rethinking Convolutional Network Design for Image Restoration (ICML 2023) · better on 2 of 2 shared results
- PromptIR — PromptIR: Prompting for All-in-One Image Restoration (NeurIPS 2023) · better on 2 of 2 shared results
- PromptRestorer — PromptRestorer: A Prompting Image Restoration Method with Degradation Perception (NeurIPS 2023) · better on 2 of 2 shared results
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 2 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
- Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration (CVPR 2026 · similarity 0.89)
- Beyond Degradation Redundancy: Contrastive Prompt Learning for All-in-One Image Restoration (TPAMI 2025 · similarity 0.89)
- Degradation-Aware Feature Perturbation for All-in-One Image Restoration (CVPR 2025 · similarity 0.89)
- Learning Where and What to Restore for Composite Image Restoration (NeurIPS 2026 · similarity 0.88)
- Parameter Efficient Adaptation for Image Restoration with Heterogeneous Mixture-of-Experts (NeurIPS 2024 · similarity 0.88)
- Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration (NeurIPS 2026 · 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.
