Research map / Image generation / Image restoration
Residual Diffusion Bridge Model for Image Restoration
Hebaixu Wang, Jing Zhang, Haoyang Chen, Haonan Guo, Di Wang, Jiayi Ma, Bo Du
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 |
|---|---|---|---|
| modulate noise with residual pi = x0 - mu instead of pi = 1 (ablation) | RDBM-L · Five-task average | PSNR: 30.15 -> 31.04 | Improved |
| modulate noise with residual instead of pi = 1 (ablation) | RDBM-L · Low-light enhancement | PSNR: 21.71 -> 24.53 | Improved |
| modulate noise with residual instead of pi = 1 (ablation) | RDBM-L · Desnowing | PSNR: 32.79 -> 32.59 | Mixed |
| use pi = 0 (flow-matching-like) instead of residual (ablation) | RDBM-L · Five-task average | PSNR: 31.04 -> 28.21 | Hurt |
| use absolute residual instead of signed residual (ablation) | RDBM-L · Five-task average | PSNR: 31.04 -> 30.94 | No clear effect |
| use cosine noise schedule instead of linear (ablation) | RDBM-L · Five-task average | PSNR: 30.99 -> 31.04 | Improved |
| use cosine noise schedule instead of linear (ablation) | RDBM-L · Desnowing | PSNR: 34.15 -> 32.59 | Hurt |
| use stationary variance lambda 10/255 instead of 100/255 (ablation) | RDBM-L · Five-task average | PSNR: 29.08 -> 31.04 | Improved |
| increase sampling steps from 10 to 100 NFE (ablation) | RDBM-L · Five-task average | PSNR: 31.04 -> 30.19 | Hurt |
| increase sampling steps from 2 to 10 NFE (ablation) | RDBM-L · Five-task average | PSNR: 22.81 -> 31.04 | Improved |
| scale RDBM from T to L variant | RDBM · Five-task average | PSNR: 27.31 -> 31.04 | Improved |
| use RDBM instead of GOUB bridge on inpainting | RDBM · CelebA-HQ inpainting | PSNR: 31.56 -> 37.88 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AutoDIR — AutoDIR: Automatic All-in-One Image Restoration with Latent Diffusion (ECCV 2024) · 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
- Image Restoration Through Generalized Ornstein-Uhlenbeck Bridge (ICML 2024 · similarity 0.93)
- IRBridge: Solving Image Restoration Bridge with Pre-trained Generative Diffusion Models (ICML 2025 · similarity 0.91)
- UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control (ICML 2025 · similarity 0.91)
- Residual Denoising Diffusion Models (CVPR 2024 · similarity 0.91)
- Diffusion Bridge Implicit Models (ICLR 2025 · similarity 0.91)
- Exploring the Design Space of Diffusion Bridge Models (NeurIPS 2025 · similarity 0.90)
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.
