- 🎓 M.S. in Electronic Information at Shanghai Jiao Tong University (2025–); B.Eng. in IntelliSense Engineering at Harbin Institute of Technology (2021–2025)
- 🔬 Interested in generative models, large language models, 3D/4D reconstruction and world models
- 🧭 Lately: how radiance fields behave when inputs are imperfect (e.g. noisy camera poses), and making evaluation metrics measure what they claim to
- 🔭 Going forward: combining generative models with 3D/4D reconstruction
- 🏆 ICM 2024 Honorable Mention · China Sensor Innovation Competition 2024, National 2nd Prize · Smart Car team captain @ HIT
| Project | Description |
|---|---|
| VGGT-Splat · page | How many frames does a casual capture need, and when do feed-forward VGGT poses beat COLMAP for 3D Gaussian Splatting? 6 scenes, 8 to all frames. |
| PoseGuard-NeRF · page | When does joint camera-pose optimization help NeRF, and when does it make the world drift? 45 controlled experiments. |
| AFVE-Bench · arXiv | When a user asks for a rewrite under a new frame, does the old one still leak through? 192 bilingual four-turn conversations; 4,204 responses from 11 model configurations, blind-annotated by two raters. |
| qingzhou | Code for the National University Students Intelligent Car Race (C++). |
| dotfiles | My macOS development setup. |
Languages
ML & Vision
3D Vision
Tools
