Ph.D. Student at HKU CDS, advised by Prof. Difan Zou.
I am a Ph.D. student in the School of Computing and Data Science (CDS) at The University of Hong Kong (HKU), where I am fortunate to be advised by Prof. Difan Zou. Prior to joining HKU, I received my B.Eng. degree in Software Engineering from Harbin Institute of Technology (HIT) (Rank 2/231, National Scholarship recipient), advised by Prof. Weigang Zhang and Prof. Shuhui Wang.
My research centers on Embodied AI and 3D Vision & Generation. I aim to build spatially grounded, physically plausible 3D world representations and generalist embodied agents capable of perceiving, reasoning, and interacting with complex physical environments. Previously, I was a research intern at MMLab@NTU (with Prof. Xingang Pan), MMLab@HKU & VAST (with Prof. Xihui Liu and Yan-Pei Cao), and Westlake University (with Prof. Huan Wang).
Equipping physical agents with goal-directed procedural planning, spatial-temporal reasoning, and generalizable action policies grounded in multimodal sensory observations.
Synthesizing part-aware, semantically decoupled, and structurally cohesive 3D assets and interactive environments that bridge neural representations with physical simulation.
Advancing diffusion and rectified flow architectures with physical constraints and latent interpolation for controllable visual synthesis and world-building.
A two-stage framework for editable, part-aware 3D object generation that couples an autoregressive 3D part bounding-box structure planner with a spatially-conditioned rectified flow generator—achieving high semantic decoupling and geometric cohesion across articulated and multi-part 3D assets.
@article{yang2025omnipart,
title={OmniPart: Part-Aware 3D Generation with Semantic Decoupling and Structural Cohesion},
author={Yang, Yunhan and Zhou, Yufan and Guo, Yuan-Chen and Zou, Zi-Xin and Huang, Yukun and Liu, Ying-Tian and Xu, Hao and Liang, Ding and Cao, Yan-Pei and Liu, Xihui},
journal={ACM Transactions on Graphics (SIGGRAPH Asia)},
year={2025}
}
Addresses goal-directed procedure planning from visual start and goal observations by introducing a latent-space temporal interpolation module and a task-adaptive masked proximity loss, enabling coherent intermediate state reasoning and state-of-the-art action sequence prediction.
@inproceedings{zhou2025masked,
title={Masked Temporal Interpolation Diffusion for Procedure Planning in Instructional Videos},
author={Zhou, Yufan and Qi, Zhaobo and Lin, Lingshuai and Jing, Junqi and Chai, Tingting and Zhang, Beichen and Wang, Shuhui and Zhang, Weigang},
booktitle={International Conference on Learning Representations (ICLR)},
year={2025}
}
A training-free concept blending framework utilizing staged latent interpolation and reverse-order feedback updates with unCLIP image conditions to synthesize visually harmonious and semantically faithful cross-concept compositions.
@article{zhou2025freeblend,
title={FreeBlend: Advancing Concept Blending with Staged Feedback-Driven Interpolation Diffusion},
author={Zhou, Yufan and Shen, Haoyu and Wang, Huan},
journal={arXiv preprint arXiv:2502.05606},
year={2025}
}Ph.D. Student · School of Computing and Data Science (CDS)
Advisor: Prof. Difan Zou
Research Focus: Embodied AI, 3D Vision & Generation
B.Eng. in Software Engineering
GPA: 3.9 / 4.0 · Rank: 2 / 231
Advisors: Prof. Weigang Zhang & Prof. Shuhui Wang
Research Intern · Advised by Prof. Xingang Pan
Explored 3D generation and geometric reconstruction pipelines integrated with 6D object pose estimation in 3D scenes.
Research Intern · Advised by Prof. Xihui Liu & Yan-Pei Cao
Investigated part-aware 3D asset generation with semantic decoupling and structural cohesion (OmniPart, SIGGRAPH Asia 2025).
Research Intern · Advised by Prof. Huan Wang
Researched generative diffusion models for concept blending, personalization, and synthetic data expansion.
A curated taxonomy and literature review of goal-directed procedure planning, action anticipation, and temporal reasoning for embodied tasks.
Comprehensive survey on leveraging generative diffusion models for dataset augmentation, representation learning, and knowledge distillation.
LLM-driven smartphone automation assistant leveraging Android Accessibility Services for cross-app task execution and UI grounding.
Interactive 3D/animated virtual digital agent integrating Retrieval-Augmented Generation (RAG) for multimodal conversational interaction.
Ministry of Education of the People's Republic of China (Top 1% nationwide)
Huawei Technologies & Harbin Institute of Technology
COMAP — Consortium for Mathematics and Its Applications
Virtual Digital Agent Project
Interactive 3D Physics Simulation with Unity3D
Harbin Institute of Technology
