Cyclone: Diffusion Model for Cycle-Consistent Weather Editing from Unpaired Driving Data

2026-07-15

AI Summary

Current autonomous driving perception methods suffer from insufficient robustness under diverse weather conditions and often rely on paired data, yielding unrealistic synthesis or limited generalization. This work proposes a unified weather editing framework based on latent diffusion models, which— for the first time—integrates cycle-consistency constraints and knowledge from image-text pretrained models into unpaired multi-weather generation and removal tasks. The approach preserves scene structure and produces photorealistic weather effects without requiring paired training data, and can be distilled into a video diffusion model to achieve temporally coherent weather editing. Experiments demonstrate that the proposed method significantly enhances performance across multiple downstream perception tasks while achieving superior visual fidelity and temporal consistency at both image and video levels.

Abstract

Reliable perception under diverse weather conditions remains a major challenge for autonomous driving systems. A common strategy to improve robustness is either to synthesize adverse weather conditions for training perception models or to apply weather-removal techniques to recover clean inputs. However, existing approaches typically rely on synthetic data augmentation or physics-based, task-specific models that require paired training data and often struggle to generate realistic weather effects or generalize robustly to out-of-domain scenarios. Toward this problem, we present Cyclone, a unified framework for weather editing based on latent diffusion, equipped with cycle-consistent constraints and knowledge from image-text models. Cyclone enables the generation of multiple weather conditions across diverse scenes while eliminating the need for paired data. Experimental results show that our approach produces more realistic, structure-preserving outputs than existing baselines and leads to consistent improvements across several downstream driving perception tasks. Furthermore, we demonstrate that Cyclone can be distilled to a video diffusion model for temporally consistent weather editing.

Problem

• autonomous driving
• weather editing
• unpaired data
• perception robustness
• domain generalization

Innovation

• latent diffusion
• cycle-consistent
• unpaired data
• weather editing
• image-text prior