Haohuan Fu

VLMs forget visual reasoning skills in continual learning because today's methods over-protect the language model while neglecting the vision encoder.

Extracting agricultural parcels from satellite imagery gets a whole lot harder (and more realistic) with a new dataset focused on the complex, irregular, and heterogeneous terrain of terraced farms.

Achieve state-of-the-art performance in multimodal remote sensing semantic segmentation with significantly fewer trainable parameters by using a novel parameter-efficient and modality-balanced symmetric fusion framework.

Injecting physics-based priors derived from MLLMs at decoding time significantly boosts weather forecasting accuracy and stability, even in long autoregressive rollouts.