Research map / Image generation / Fast diffusion sampling
Tuning Real-World Image Restoration at Inference: A Test-Time Scaling Paradigm for Flow Matching Models
Purui Bai, Junxian Duan, Pin Wang, Jinhua Hao, Ming Sun, Chao Zhou, Huaibo Huang
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 VLM text caption condition (ablation) | ResFlow-Tuner · LSDIR-Val | LPIPS: 0.3856 -> 0.3721 | Improved |
| use UMMF unified-sequence fusion instead of direct adding (ablation) | ResFlow-Tuner · LSDIR-Val | FID: 28.39 -> 19.82 | Improved |
| use UMMF instead of ControlNet (ablation) | ResFlow-Tuner · LSDIR-Val | FID: 24.73 -> 19.82 | Improved |
| use three-reward verifier ensemble instead of Aesthetic alone (ablation) | ResFlow-Tuner · LSDIR-Val | MANIQA: 0.6224 -> 0.6658 | Improved |
| use verifier ensemble instead of CLIPScore + ImageReward pair (ablation) | ResFlow-Tuner · LSDIR-Val | MUSIQ: 70.98 -> 74.21 | Improved |
| add test-time scaling (K=4, N=7) (ablation) | ResFlow-Tuner · LSDIR-Val | FID: 22.91 -> 19.82 | Improved |
| increase TTS to K=10, N=15 (ablation) | ResFlow-Tuner · LSDIR-Val | FID: 19.82 -> 18.17 | Improved |
| increase TTS to K=10, N=15 (compute cost) (ablation) | ResFlow-Tuner · LSDIR-Val | inference compute: 0.375e4 -> 1.0e4 | Hurt |
| use ResFlow-Tuner versus GAN baselines on full-reference fidelity | ResFlow-Tuner · DIV2K-Val-I | PSNR: 27.63 -> 23.11 | Hurt |
| use ResFlow-Tuner versus FaithDiff on no-reference quality | ResFlow-Tuner · DIV2K-Val-I | MUSIQ: 66.53 -> 75.54 | Improved |
| restore with ResFlow-Tuner before OCR | ResFlow-Tuner · Occluded RoadText 2024 | OCR precision (%): 36.45% -> 42.23% | 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.
- FaithDiff — FaithDiff: Unleashing Diffusion Priors for Faithful Image Super-resolution (CVPR 2025) · better on 7 of 7 shared results
- SeeSR — SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 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
- Reversing Flow for Image Restoration (CVPR 2025 · similarity 0.90)
- Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild (CVPR 2024 · similarity 0.90)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.89)
- DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation (NeurIPS 2024 · similarity 0.89)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.89)
- Posterior-Mean Rectified Flow: Towards Minimum MSE Photo-Realistic Image Restoration (ICLR 2025 · 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.
