Research map / Image generation / Image restoration
FoundIR: Unleashing Million-scale Training Data to Advance Foundation Models for Image Restoration
Hao Li, Xiang Chen, Jiangxin Dong, Jinhui Tang, Jinshan Pan
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 |
|---|---|---|---|
| use incremental learning from isolated to coupled degradations instead of Mix-Train (ablation) | FoundIR generalist · FoundIR benchmark | PSNR: 22.57 -> 23.70 | Improved |
| use incremental learning instead of Combine-Train (ablation) | FoundIR generalist · FoundIR benchmark | PSNR: 23.40 -> 23.70 | Improved |
| use incremental learning instead of Sequence-Train (ablation) | FoundIR generalist · FoundIR benchmark | PSNR: 22.71 -> 23.70 | Improved |
| order incremental learning isolated first then coupled rather than coupled then isolated (ablation) | FoundIR generalist · FoundIR benchmark | PSNR: 22.33 -> 23.70 | Improved |
| add specialist models after the generalist (G -> S) versus generalist only (ablation) | FoundIR · FoundIR benchmark L+H | PSNR: 18.19 -> 22.82 | Improved |
| add generalist before specialist versus specialist only (ablation) | FoundIR · FoundIR benchmark L+H | PSNR: 13.06 -> 22.82 | Improved |
| apply specialist first then generalist (S -> G) instead of G -> S (ablation) | FoundIR · FoundIR benchmark L+H | PSNR: 22.82 -> 15.63 | Hurt |
| chain two specialists (S1 -> S2) instead of generalist then specialist (ablation) | FoundIR · FoundIR benchmark L+H | PSNR: 22.82 -> 13.52 | Hurt |
| train on 10% of the proposed dataset instead of 120,000 public real-world pairs (ablation) | FoundIR · FoundIR benchmark | PSNR: 2.59dB improvement in PSNR | Improved |
| increase training data from 10% to 100% of the proposed dataset (ablation) | FoundIR · FoundIR benchmark | PSNR | Improved |
| set degradation-agnostic coefficient gamma_T to 0.3 versus other values or none (ablation) | FoundIR generalist · FoundIR benchmark (20 degradation types) | PSNR | Improved |
| train DiffUIR on the proposed dataset instead of public datasets | DiffUIR · RealRain-1K | PSNR: 6dB PSNR gain | Improved |
Later papers that compared with this method
3 later accepted papers report FoundIR as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- FoundIR-v2: Optimizing Pre-Training Data Mixtures for Image Restoration Foundation Model (CVPR 2026 · better on 3 of 3 shared results)
- Expert Weaving: Marrying Masked AutoRegressive and Diffusion Models for Unified Image Restoration (ECCV 2026 · better on 3 of 3 shared results)
- MMDIR: Multimodal Instruction-Driven Framework for Mixed-Degradation Document Image Restoration (CVPR 2026 · better on 2 of 2 shared results)
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 1 of 1 shared result
- 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
- DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation (NeurIPS 2024 · similarity 0.90)
- ReFIR: Grounding Large Restoration Models with Retrieval Augmentation (NeurIPS 2024 · similarity 0.89)
- GenDeg: Diffusion-based Degradation Synthesis for Generalizable All-In-One Image Restoration (CVPR 2025 · similarity 0.88)
- Parameter Efficient Adaptation for Image Restoration with Heterogeneous Mixture-of-Experts (NeurIPS 2024 · similarity 0.88)
- Learning from History: Task-agnostic Model Contrastive Learning for Image Restoration (AAAI 2024 · similarity 0.88)
- IRPO: Boosting Image Restoration via Post-training GRPO (NeurIPS 2026 · 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.
