Shangzhe Di   狄尚哲

Hi, I am a fourth-year PhD candidate at Shanghai Jiao Tong University (SJTU), where I am fortunate to be advised by Prof. Weidi Xie. My research focuses on video understanding and multimodal learning, driven by a passion for exploring the unknowns in these fields.

Before joining SJTU, I earned my master's and bachelor's degrees from Beihang University. During this period, I explored video background music generation under the guidance of Prof. Si Liu.

I have also gained industry experience through internships at Alibaba and ByteDance Seed. I am currently a Student Researcher at Google DeepMind (FAI), where I work on 4D world models and dynamic spatial memory.

I'm graduating in 2027 and actively seeking opportunities in academia and industry. My research focuses on video temporal grounding, streaming video understanding, and visual representation learning — feel free to reach out if you think there's a good fit!

Email  /  CV  /  Github  /  Google Scholar

Multitask pre-training can yield scalable general-purpose visual representations.

Towards universal video grounding with superior accuracy, generalizability, and robustness.

Learn streaming video representations of various granularities via multitask training, including global (retrieval, action recognition), temporal (grounding), and spatial (segmentation) objectives.

Distill multi-step reasoning and spatial-temporal understanding into a generative Video-LLM.

A training-free approach enabling Video-LLMs for streaming video question-answering.

Pinpoint scattered visual evidence in long egocentric videos while responding to questions.

Simultaneous query grounding and answering in long, egocentric videos.

The first satisfying method for video background music generation.