Exploring Weather-aware Aggregation and Adaptation for Semantic Segmentation under Adverse Conditions

Semantic segmentation under adverse conditions is critical for reliable visual perception in challenging weather environments. These extreme scenarios introduce distortions, such as low contrast and reduced visibility, making traditional segmentation models struggle. The scarcity of labeled data in such conditions makes it difficult to train models directly for these environments. Unsupervised domain adaptation (UDA) has been proposed as a solution to transfer knowledge from labeled source domains (normal weather) to unlabeled target domains (adverse weather). However, existing methods face significant challenges, particularly due to weather unawareness and feature heterogeneity. Many models fail to account for the unique characteristics of different weather conditions, and the significant feature discrepancies between normal and adverse weather images hinder effective adaptation. In this paper, we propose a novel weather-aware aggregation and adaptation network that leverages characteristic knowledge to achieve weather homogenization and enhance scene perception. Specifically, we introduce amplitude prompt aggregation to capture essential characteristics from the Fourier frequency domain that are indicative of different weather conditions. Additionally, we employ weather heterogeneity adaptation to mitigate the inter-domain heterogeneity, thereby achieving feature homogenization across diverse environments. Extensive experimental results on multiple challenging benchmarks demonstrate that our method achieves consistent improvements for semantic segmentation under adverse conditions.