Jiaxin Liu

Event-VLA achieves robust robotic manipulation even in near-dark conditions by effectively integrating motion-sensitive event streams with traditional RGB inputs.

Universal image editing models are surprisingly competitive with specialized virtual try-on systems, but still stumble when handling multiple garments simultaneously.

Diffusion models can learn surprisingly generalizable anatomical representations from unlabeled MRI data, enabling accurate multi-task diagnosis and segmentation across different joints and imaging conditions, even with limited labeled data.