Research Fellow at Nanyang Technological University, working with Assoc. Prof. Kang Hao Cheong.
I study the dynamic states of humans and AI in open environments: what a system should keep, update, or leave open as new observations arrive. I design tasks and benchmarks that examine these states directly, build models that hold them better, and test them in open-environment visual perception, long and streaming video understanding, and human–AI education.
Homepage · Google Scholar · CV · LinkedIn · Email
I build and maintain public platforms for evaluating dynamic vision.
| Platform | Purpose |
|---|---|
| VideoCube / MGIT | Global instance tracking and its multi-modal extension (TPAMI 2023, NeurIPS 2023). Toolkit |
| SOTVerse / VLTVerse | User-defined task spaces for tracking (IJCV 2024) |
| BioDrone | Tracking from a bionic drone under physical disturbance (IJCV 2024). Toolkit |
| GOT-10k | Generic object tracking benchmark (TPAMI 2021); I maintain the platform |
- ResearchOS: a local research memory in plain files, for keeping your own judgment across AI-assisted work.
- FIOVA: a multi-annotator benchmark for human-aligned video captioning.
- Studying the Human Mind with AI: a reading companion to my talk, with guides, key readings and a glossary in English and Chinese.
- Ph.D., Institute of Automation, Chinese Academy of Sciences, 2024. Supervised by Prof. Kaiqi Huang and co-supervised by Prof. Xin Zhao.
- M.Sc. in Computer Science, The University of Hong Kong, 2019. Supervised by Prof. Cho-Li Wang.
- B.E. in Information Engineering, Beijing Institute of Technology, 2017.
I welcome research discussions and collaborations. Please reach me at shiyu.hu@ntu.edu.sg.