International Conference on Machine Learning (ICML) · 2026
AI · 3D vision · human motion
Nanjie
Yao.
Ph.D. student in Artificial Intelligence
HKUST (Guangzhou)
I study how to recover human shape, appearance, and motion from limited observations—and how intelligent systems can reason and act on what they see.

01 / Background
About me.
I joined Prof. Hao Wang’s group at the Hong Kong University of Science and Technology (Guangzhou) as a research assistant in April 2024. My early projects focused on reconstructing a 3D person from a single image. The questions gradually moved from shape and texture to animation and motion, and I stayed for a Ph.D. in Artificial Intelligence. I still spend most of my time on 3D human reconstruction and motion recovery, especially when the observations are sparse: one image, a few views, or signals from a head-mounted device.
Alongside that work, I am exploring LLM agents and how they can coordinate with specialized vision and action models. I am currently an Algorithm Engineer Intern at Horizon Robotics, where I work on video generation, world models, and vision-language-action research. I also worked remotely with Prof. Zhengzhong Tu at Texas A&M University on human video generation.
Before coming to HKUST (Guangzhou), I studied Network Engineering at Zhejiang University of Technology with Prof. Sixian Chan, and graduated from Jianxing Honors College. I spent a term at the University of California, Irvine, worked on 3D reconstruction at Zhejiang Lab, and explored Auto-ML and neural backdoor mitigation during my undergraduate years.
02 / Research
Selected work.
IEEE Transactions on Visualization and Computer Graphics (IEEE TVCG) · 2026
MultiGO++: Monocular 3D Clothed Human Reconstruction via Geometry-Texture Collaboration
Annual AAAI Conference on Artificial Intelligence · 2026
FastAnimate: Towards Learnable Template Construction and Pose Deformation for Fast 3D Human Avatar Animation
Computer Vision and Pattern Recognition (CVPR) · 2025
MultiGO: Towards Multi-level Geometry Learning for Monocular 3D Textured Human Reconstruction
IEEE Transactions on Dependable and Secure Computing (IEEE TDSC) · Accepted · 2026
DeepSanitizer: Data-free Backdoor Removal in Deep Neural Networks
ACM Computing Surveys (ACM CSUR) · Accepted · 2026
Orchestrating LLMs with Specialized Models: A Survey on Heterogeneous Multi-Model Agents
03 / Contact