Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

P-Flow: Proxy-gradient Flows for Linear Inverse Problems

About

Generative models based on flow matching have emerged as a powerful paradigm for inverse problems, offering straighter trajectories and faster sampling compared to diffusion models. However, existing approaches often necessitate differentiating through unrolled paths, leading to numerical instability and prohibitive computational overhead. To address this, we propose P-Flow, a framework that stabilizes the reconstruction process by leveraging a proxy gradient to update the source point. This approach effectively circumvents the numerical instability and memory overhead of long-chain differentiation. To ensure consistency with the prior distribution, we employ a Gaussian spherical projection motivated by the concentration of measure phenomenon in high-dimensional spaces. We further provide a theoretical analysis for P-Flow based on Bayesian theory and Lipschitz continuity. Experiments across diverse restoration tasks demonstrate that P-Flow delivers competitive performance, especially under extreme degradations such as severely ill-posed conditions and high measurement noise.

Zehua Jiang, Fenghao Zhu, Xinquan Wang, Chongwen Huang, Zhaoyang Zhang• 2026

Related benchmarks

TaskDatasetResultRank
Image InpaintingFFHQ (test)
LPIPS0.154
73
Super-ResolutionFFHQ (test)
SSIM62
32
DenoisingAFHQ Cat (test)
PSNR32.09
30
Super-ResolutionAFHQ Cat (test)
LPIPS0.157
26
Random InpaintingAFHQ Cat (test)
PSNR32.73
24
Box InpaintingAFHQ Cat (test)
PSNR27.56
20
Super-ResolutionFFHQ 128 images (test)
PSNR27.64
9
DenoisingFFHQ 128 (test)
PSNR32.45
9
Random InpaintingFFHQ 128 images (test)
PSNR33.29
9
DeblurringFFHQ 128 images (test)
PSNR28.07
9
Showing 10 of 12 rows

Other info

Follow for update