Plug-and-Play ADMM for Inverse Problems with Flow Matching Denoiser
Organizations: TH Würzburg-Schweinfurt, CAIRO · University of Würzburg
Abstract
Plug-and-Play (PnP) methods with generative priors have achieved strong performance in inverse problems, with recent flow matching models enabling fast and scalable reconstruction. However, a key challenge in PnP methods is maintaining consistency between data fidelity updates and the learned generative prior, as data-consistency steps can move iterates away from the flow trajectory. We propose an ADMM-based PnP framework for flow matching priors that enforces stronger consistency between the measurement model and the prior through auxiliary variables and dual updates. In addition, we introduce a trajectory correction mechanism that restores consistency with the flow via a combination of deterministic and stochastic projections. Our method provides a more principled integration of flow priors within PnP frameworks, achieving high-quality reconstructions with reduced computational cost.
Figures & tables
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
| AFHQ-CAT | CelebA | |||||
| Method | SSIM | LPIPS | PSNR | SSIM | LPIPS | PSNR |
| PNP-ADMM (adaptive) | 0.74 | 0.16 | 26.26 | 0.80 | 0.07 | 27.55 |
| PNP-ADMM (random) | 0.71 | 0.21 | 25.81 | 0.79 | 0.09 | 26.47 |
| PNP-ADMM (deterministic) | 0.61 | 0.23 | 23.78 | 0.73 | 0.10 | 26.16 |
| PNP-FBS | 0.70 | 0.36 | 25.15 | 0.79 | 0.15 | 26.18 |
| AFHQ-Cat | CelebA | |||||
| SSIM | LPIPS | PSNR | SSIM | LPIPS | PSNR | |
| 5 | 0.51 | 0.41 | 24.91 | 0.55 | 0.31 | 24.26 |
| 10 | 0.56 | 0.38 | 25.73 | 0.60 | 0.29 | 24.85 |
| 15 | 0.60 | 0.36 | 26.13 | 0.62 | 0.28 | 25.04 |
| 20 | 0.54 | 0.39 | 25.29 | 0.58 | 0.30 | 24.45 |
| PnP steps | Early Stop | AFHQ-Cat | CelebA | ||||
| SSIM | LPIPS | PSNR | SSIM | LPIPS | PSNR | ||
| 200 | 170 | 0.58 | 0.30 | 19.05 | 0.60 | 0.19 | 17.58 |
| 300 | 255 | 0.65 | 0.22 | 21.68 | 0.67 | 0.14 | 19.76 |
| 400 | 340 | 0.68 | 0.19 | 23.63 | 0.66 | 0.12 | 23.51 |
| 800 | 680 | 0.72 | 0.13 | 27.47 | 0.76 | 0.07 | 26.01 |
| 1000 | 850 | 0.72 | 0.12 | 28.05 | 0.76 | 0.06 | 26.99 |
Inverse Problems
Inverse problems infer hidden causes, parameters, or signals from indirect, noisy measurements using physical models, priors, and statistical methods.
See all 262 papers