2025 - Present
Founder - Stealth Startup
I’m building creation tools powered by the latest AI models.
Di Huang
Startup Founder
Hey, I’m Di. Thanks for stopping by. I write here about what I’m exploring, building, and learning along the way.
We are currently building a tiny startup with a group of curious people. Our goal is to make creative work an order of magnitude easier, so anyone can turn their ideas into real things. What once required rare skills, expensive tools, or a full production team can now start from a single idea.
Experience
2022 - 2025
Ph.D. - The University of Sydney
My Ph.D. focused on deep learning for 3D vision. I was advised by Wanli Ouyang and worked closely with Tong He.
In the beginning, my research explored how better 3D priors and network designs could help solve 3D problems. Later, as NLP advanced rapidly and scaling models and compute started to show remarkable results, I became interested in bringing similar ideas into 3D vision. In particular, I worked on 3D pre-training and explored how to train larger models for 3D understanding and reconstruction.
Along the way, driven mostly by curiosity, I also worked with many junior students on a range of interesting topics, including AI agents and video generation.
I loved my time in Sydney. It is a beautiful city.
2016 - 2020
B.E. - Zhejiang University
Publications
During my Ph.D., I spent several wonderful years exploring how AI can understand the physical world: space, shape, and motion. My research moved across embodied models (Ponder, PonderV2, UniPAD, Agent3D-Zero, Point Cloud Matters), 3D reconstruction (NeuRodin, MeshAnything, GVGEN, HHOR, OnePose++), and video understanding (Depth Any Video, MotionGPT, Holistic-Motion2D, Sentry, PredBench, FiT, EasyMoCap).