Beibei Lin

Nighttime video deraining models can now generalize to the real world, thanks to a massive, physically-grounded synthetic dataset that finally captures the complex interplay of rain and artificial light.

VLMs struggle to reason about visual scenes in adverse weather, losing significant segmentation accuracy as rain, snow, or fog intensifies.

Despite advances in vision-language models, reasoning across sparse, multi-view observations remains surprisingly unsolved, with current models barely outperforming random guessing on a new benchmark.