Common weather phenomena including rain, snow, fog, and haze introduce visual degradations that significantly impact the performance of semantic segmentation algorithms. These degradations include partial to severe occlusions of objects, illumination changes, noise, and domain shift. As most vision algorithms assume clear weather conditions, their performance suffers from the domain gap between clean and degraded imagery.

UG²+ Track 2 aims to spark the development of novel semantic segmentation algorithms for images captured under adverse weather conditions. Participants are challenged to develop models that are robust to the full range of weather degradations encountered in real-world deployments.

Full competition details, dataset downloads, submission instructions, and evaluation guidelines are available on the official competition page on Codabench:

View Competition on Codabench

If you have any questions about this challenge track please feel free to email cvpr2026.ug2challenge@gmail.com