Perception in Adverse Weather (Qian, University of Leeds)

Abstract

This research explores how adverse weather can be modeled for autonomous systems and how robustness can be improved under such conditions. Real-world data collection in rain, snow, or fog is limited, and existing datasets often lack fine-grained physical detail. To address this, a unified weather-removal framework is developed to enhance perception robustness, alongside a controllable simulation framework that models atmospheric effects within 3D scene representations. This enables more realistic training conditions and supports applications in urban-scale reconstruction and digital twins under changing environments.

Chenghao Qian is currently a PhD candidate at the University of Leeds, working on generative spatial modelling of adverse weather. Prior to this, he completed his master’s degree at the University of Sydney and gained several years of experience in robotics and autonomous driving at Parallel Domain, XPENG, and UBTech. His overarching goal is to enable robots to operate reliably across any location, under any weather conditions, at any time. His research interests include 3D vision, robotics, and autonomous driving. He has published papers at top-tier conferences such as ICCV, AAAI and IROS. He also organizes CVPR 2026 LoViF Workshop.

Feel free to reach out to explore potential collaborations! Stay tuned for more from the CV4DT and CAMCV!