Jisheng Dang

Injecting physics-inspired spatial priors into video MLLMs drastically enhances temporal stability and segmentation accuracy without sacrificing image-level performance.

CARE transforms the approach to reasoning length in video-MLLMs, enabling models to adaptively balance exploration and efficiency based on their evolving competence.

VideoCFR not only boosts performance in video reasoning tasks but also reveals the critical visual evidence driving model decisions without relying on human annotations.

State-of-the-art performance in social intelligence reasoning is achieved by ensuring long-tail events are prioritized over head events through innovative knowledge distillation techniques.