WRD:RGB-IR dual-spectral dataset for road surface classification in severe winter

The winter road-surface classification dataset (WRD) is a large-scale dual-spectral dataset designed to advance autonomous driving perception in adverse winter conditions. Collected in Harbin, China, it contains approximately 200,000 synchronized RGB and thermal infrared image pairs with comprehensive annotations covering diverse road surface states, including dry, wet, and snow-covered conditions. Designed to facilitate the development of robust all-weather perception systems, WRD flexibly supports the evaluation and validation of both single-modal and dual-modal algorithms. Furthermore, pre-trained weights from its classification tasks can provide robust initialization in downstream applications.