Restoration-Aligned Generative Flow Models for Blind Motion Deblurring

2026-05-09

AI Summary

Existing generative flow models for blind motion deblurring suffer from reduced fidelity due to a mismatch between their training objective and the actual restoration goal. To address this, this work proposes DeblurFlow, a novel framework that aligns the vector field with the residual between blurred and sharp images by replacing the noisy endpoints of flow trajectories with the observed blurry image. The method introduces an r-space residual latent representation and a dual-expert sampling strategy, achieving both high perceptual quality and fidelity. By integrating LoRA-based fine-tuning of a pretrained flow model with a VAE architecture, DeblurFlow significantly outperforms existing baselines across four benchmarks, including GoPro. It achieves an initial PSNR of 33.69 dB, which only slightly decreases to 33.05 dB after enhancement—substantially surpassing the baseline’s 27.60 dB—while reducing computational cost by up to 9×.

Abstract

Generative flow models offer powerful priors learned from large-scale natural images, but directly adapting them to restoration tasks such as motion deblurring causes severe fidelity degradation, as their training objective is inherently misaligned with restoration. We present DeblurFlow, a framework that resolves this misalignment by reformulating the flow trajectory itself: we replace the noise endpoint with the blur observation, which makes the underlying vector field coincide with the residual error between blur and clean images. Under this formulation, the standard flow matching loss naturally takes the form of a residual loss, allowing pretrained flow models to be optimized under restoration-aligned objectives via LoRA adaptation. This formulation further enables a dual-expert sampling strategy: a fidelity expert provides a high-fidelity initialization, e.g., PSNR 33.69 dB, and DeblurFlow enhances perceptual quality with only a marginal fidelity reduction to 33.05 dB, whereas directly applying a generative model on top of a fidelity expert decreases PSNR to 27.60 dB. To make this practical, we further introduce r-space, a latent space tailored for residual decoding rather than image reconstruction, which reduces encoder-decoder cost by up to 9$\times$over standard VAE latents. Extensive experiments on GoPro, HIDE, RealBlur, and RWBI demonstrate that DeblurFlow achieves strong restoration fidelity and perceptual realism, while remaining computationally practical.

Problem

• blind motion deblurring
• generative flow models
• fidelity degradation
• restoration alignment
• image restoration

Innovation

• Generative Flow Models
• Blind Motion Deblurring
• Restoration-Aligned Objective
• Residual Decoding
• LoRA Adaptation