Research map / Image generation / Image restoration
Learning Continuous Wasserstein Barycenter Space for Generalized All-in-One Image Restoration
Xiaole Tang, Xiaoyi He, Jiayi Xu, Xiang Gu, Jiayu Sun
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 Wasserstein barycenter learning with PromptIR backbone | PromptIR · 3 degradations average | PSNR: 32.05 -> 33.13 | Improved |
| add Wasserstein barycenter learning with Restormer backbone | Restormer · 5 degradations average | PSNR: 27.46 -> 30.72 | Improved |
| add Wasserstein barycenter learning with PromptIR backbone | PromptIR · 5 degradations average | PSNR: 29.72 -> 31.05 | Improved |
| use WB embeddings instead of original embeddings (ablation) | BaryIR · 5 degradations average | PSNR: 27.46 -> 29.89 | Improved |
| add residual embeddings to WB embeddings (ablation) | BaryIR · 5 degradations average | PSNR: 29.89 -> 30.72 | Improved |
| add residual embeddings to original embeddings (no WB) (ablation) | Restormer · O-HAZE | PSNR: 18.02 -> 19.84 | Improved |
| add all three losses MWB, IRC, BRO (ablation) | BaryIR · 5 degradations average | PSNR: 27.72 -> 30.72 | Improved |
| add MWB loss alone (ablation) | BaryIR · 5 degradations average | PSNR: 27.72 -> 29.37 | Improved |
| use transformer-block transport map instead of MLP (ablation) | BaryIR · 5 degradations average | PSNR: 30.04 -> 30.72 | Improved |
| use portion-based barycenter weights instead of uniform (ablation) | BaryIR · 3 degradations average | PSNR: 32.67 -> 32.86 | Improved |
| increase training batch size beyond 8 (ablation) | BaryIR · 5 degradations | PSNR | No clear effect |
| apply BaryIR to unseen JPEG artifacts | BaryIR vs MoCE-IR · BSD500 QF=10 | PSNR: 26.42 -> 27.94 | Improved |
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.
- AdaIR — AdaIR: Adaptive All-in-One Image Restoration via Frequency Mining and Modulation (ICLR 2025) · 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
- 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
- Retrieve-to-Restore: Efficient All-in-One Image Restoration with a Retrieval-Based Degradation Bank (CVPR 2026 · similarity 0.90)
- Degradation-Aware Feature Perturbation for All-in-One Image Restoration (CVPR 2025 · similarity 0.87)
- Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration (NeurIPS 2026 · similarity 0.87)
- GenDeg: Diffusion-based Degradation Synthesis for Generalizable All-In-One Image Restoration (CVPR 2025 · similarity 0.86)
- Learning Where and What to Restore for Composite Image Restoration (NeurIPS 2026 · similarity 0.86)
- Learning from History: Task-agnostic Model Contrastive Learning for Image Restoration (AAAI 2024 · similarity 0.86)
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.
