Research map / Image generation / Fast diffusion sampling
Saving Foundation Flow-Matching Priors for Inverse Problems
Yuxiang Wan, Ryan Devera, Wenjie Zhang, Ju 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 time-dependent warm-start (ablation) | FMPlug · DIV2K Gaussian deblur | PSNR (dB): 25.1602 -> 26.0547 | Improved |
| add sharp Gaussianity regularization on top of warm-start (ablation) | FMPlug · DIV2K Gaussian deblur | PSNR (dB): 26.0547 -> 26.2563 | Improved |
| add sharp Gaussianity regularization on top of warm-start (ablation) | FMPlug · DIV2K Gaussian deblur | LPIPS: 0.4315 -> 0.4120 | Improved |
| use FMPlug instead of D-Flow | D-Flow · AFHQ Gaussian deblur | PSNR (dB): 25.90 -> 30.41 | Improved |
| use FMPlug instead of DIP on random inpainting | DIP · AFHQ 70% random inpainting | PSNR (dB): 33.32 -> 32.81 | Hurt |
| use FMPlug instead of FlowDPS-P | FlowDPS-P · AFHQ 4x super-resolution | PSNR (dB): 28.75 -> 30.31 | Improved |
| use FMPlug with few-shot guidance instead of D-Flow | D-Flow · InverseBench LIS | PSNR (dB): 19.67 -> 34.65 | Improved |
| use FMPlug instead of DIP | DIP · InverseBench BKH | PSNR (dB): 17.63 -> 22.91 | Improved |
| use foundation FM prior plus text prompt and inversion-seed strengthening techniques | FlowDPS and D-Flow · Gaussian deblurring AFHQ-Cat | quality | No clear effect |
| set shell thickness epsilon to 0.025 (ablation) | FMPlug · DIV2K Gaussian deblur | PSNR (dB): 26.45 -> 26.58 | Improved |
| increase NFE from 1 to 5 (ablation) | FMPlug · DIV2K Gaussian deblur | PSNR (dB): 26.44 -> 26.62 | No clear effect |
| increase blur level sigma from 1 to 6 (ablation) | FMPlug · DIV2K Gaussian deblur | learned warm-start time t*: 0.83 -> 0.72 | Hurt |
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.
- D-Flow — DFlow: A Generative Model Combining Denoising AutoEncoder and Normalizing Flow for High Fidelity Waveform Generation (ICML 2024) · better on 7 of 7 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
- Solving Inverse Problems with FLAIR (NeurIPS 2025 · similarity 0.88)
- Latent Refinement via Flow Matching for Training-free Linear Inverse Problem Solving (NeurIPS 2025 · similarity 0.87)
- InverseBench: Benchmarking Plug-and-Play Diffusion Priors for Inverse Problems in Physical Sciences (ICLR 2025 · similarity 0.87)
- Exact Posterior Score Estimation for Solving Linear Inverse Problems (NeurIPS 2026 · similarity 0.87)
- Fast samplers for Inverse Problems in Iterative Refinement models (NeurIPS 2024 · similarity 0.87)
- PnP-Flow: Plug-and-Play Image Restoration with Flow Matching (ICLR 2025 · 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.
