PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow

2026-05-13

AI Summary

This work addresses the challenge of image degradation under complex and unknown adverse weather conditions in real-world scenarios, where conventional deraining or dehazing methods often produce overly smoothed results. The authors propose a unified framework that first leverages a frozen vision-language model to enable zero-shot, soft weather awareness (via the AWR-QA module) for estimating both weather type and low-level attributes. These estimates then guide a restoration network through Attribute Modulation Normalization (AMN) and a Weather-Weighted Adapter (WWA) to generate an initial output. Finally, a terminal-consistent velocity-parameterized residual correction flow refines the result. The method significantly outperforms existing approaches on both single-type and composite weather removal tasks, achieving superior performance in terms of fidelity, perceptual quality, and cross-dataset generalization.

Abstract

Adverse weather removal (AWR) in real-world images remains challenging due to heterogeneous and unseen degradations, while distortion-driven training often yields overly smooth results. We propose PVRF, a unified framework that integrates zero-shot soft weather perceptions with velocity-constrained rectified-flow refinement. PVRF introduces an AWR-specific question answering module (AWR-QA) that uses frozen vision--language models (VLMs) to estimate soft probabilities of weather types and low-level attribute scores. These perceptions condition restoration networks via attribute-modulated normalization (AMN) and weather-weighted adapters (WWA), producing an anchor estimate for refinement. We then learn a terminal-consistent residual rectified flow with perception-adaptive source perturbation and a terminal-consistent velocity parameterization to stabilize learning near the terminal regime. Extensive experiments show that PVRF improves both fidelity and perceptual quality over state-of-the-art baselines, with strong cross-dataset generalization on single and combined degradations. Code will be released at https://github.com/dongw22/PVRF.

Problem

• adverse weather removal
• heterogeneous degradations
• unseen degradations
• over-smooth results
• real-world images

Innovation

• rectified flow
• vision-language model
• adverse weather removal
• attribute-modulated normalization
• zero-shot perception