Jun Zhang

TimeLens: Rethinking Video Temporal Grounding with Multimodal LLMs

p-MoD: building mixture-of-depths MLLMs via progressive ratio decay