LI Heyuan 李贺元

I am a Ph.D. student at The Chinese University of Hong Kong, Shenzhen, working on 3D computer vision and graphics, advised by Prof. HAN Xiaoguang. Previously, I obtained my M.S. from National University of Singapore, advised by Prof. Robby T. Tan. I received my Bachelor's degree from University of Electronic Science and Technology of China.

I believe simple is better than complex.

I (try to) practice Slow Science. Slow is faster than fast.

I expect to graduate in June 2027 and am actively seeking full-time research opportunities in 3D generation and related areas. Please feel free to get in touch.

Email / WeChat / Github

Research

My research focuses on 3D generation and reconstruction. At Tencent Hunyuan3D, I contribute to the development of native 3D generative foundation models, focusing on ultra-high-fidelity geometry generation. My prior work in human-centric 3D vision and graphics spans neural 3D representations, reconstruction, and controllable generation, with an emphasis on generalization and learning from imperfect or synthetic data at scale. I am also exploring agentic 3D modeling, with broader interests in world models and embodied intelligence.

Publications

( ^* Equal contribution. ^✉ Corresponding author. Representative papers. )

	LHM++: An Efficient Large Human Reconstruction Model for Pose-free Images to 3D Lingteng Qiu^, Peihao Li^, Heyuan Li^, Qi Zuo, Xiaodong Gu, Yuan Dong, Weihao Yuan, Rui Peng, Siyu Zhu, Xiaoguang Han, Guanying Chen^✉, Zilong Dong^✉ In Submission*, 2026 project page / arXiv Introduce an efficient Point-Image Transformer that fuses hierarchical 3D point features with multimodal attention to generate high-fidelity, animatable 3D humans from pose-free images.
	ViSA: 3D-Aware Video Shading for Real-Time Upper-Body Avatar Creation Fan Yang^, Heyuan Li^, Peihao Li, Weihao Yuan, Lingteng Qiu, Chaoyue Song, Cheng Chen, Yisheng He, Shifeng Zhang^✉, Xiaoguang Han, Steven Hoi, Guosheng Lin^✉ In Submission, 2026 project page / arXiv Combine robust 3D reconstruction priors with real-time autoregressive video diffusion to generate high-fidelity, dynamic upper-body avatars.
	Condition Matters in Full-head 3D GANs Heyuan Li^, Huimin Zhang^, Yuda Qiu, Zhengwentai Sun, Keru Zheng, Lingteng Qiu, Peihao Li, Qi Zuo, Ce Chen, Yujian Zheng, Yuming Gu, Zilong Dong^✉, Xiaoguang Han^✉ ICLR, 2026 project page / arXiv / code Enable view-invariant generation through front-view-aligned semantic features and a large-scale, balanced synthetic dataset.
	HyPlaneHead: Rethinking Tri-plane-like Representations in Full-Head Image Synthesis Heyuan Li, Kenkun Liu, Lingteng Qiu, Qi Zuo, Keru Zheng, Zilong Dong, Xiaoguang Han NeurIPS, 2025 project page / arXiv / code Propose a hy-plane representation that combines the strengths of tri-plane and spherical tri-plane methods while overcoming their inherent limitations.
	DiffPortrait360: Consistent Portrait Diffusion for 360 View Synthesis Yuming Gu, Phong Tran, Yujian Zheng, Hongyi Xu, Heyuan Li, Adilbek Karmanov, Hao Li CVPR, 2025 project page / arXiv / code Integrate a custom ControlNet and dual appearance module to generate globally consistent 360-degree head views capable of high-quality free-viewpoint rendering for both realistic and stylized head images.
	SphereHead: Stable 3D Full-head Synthesis with Spherical Tri-plane Representation Heyuan Li, Ce Chen, Tianhao Shi, Yuda Qiu, Sizhe An, Guanying Chen, Xiaoguang Han ECCV, 2024 (Oral Presentation, top 2.3%) project page / arXiv / code Achieve 3D full-head synthesis without mirroring-face and multiple-face artifacts via spherical tri-plane representation and view-image consistency loss.
	Zero-shot Real Facial Attribute Separation and Transfer at Novel Views Dingyun Zhang, Heyuan Li, Juyong Zhang CVM, 2024 paper A model that enables real-time and zero-shot attribute separation of a given real face, allowing attribute transfer and rendering at novel views without the aid of multi-view information.
	DSFNet: Dual Space Fusion Network for Occlusion-Robust 3D Dense Face Alignment Heyuan Li, Bo Wang, Yu Cheng, Mohan Kankanhalli, Robby T. Tan CVPR, 2023 project page / arXiv / video / code Infer 3D face reconstruction in both image space and model space to achieve high robustness and accuracy.

Earlier Projects

	Detailed 3D Face Reconstruction Reconstruct a detailed 3D face from a single image through self-supervised learning and differentiable rendering-based optimization.
	Cloth Simulation via Deep Learning Use neural networks to represent cloth, external objects, and the environment, and conduct a learning-based cloth simulation.
	Vehicle Trajectory Forecasting Trajectory Forecasting with Neural Networks: An Empirical Evaluation and A New Hybrid Model TITS, 2020 Conduct the most comprehensive evaluation of various models proposed for time series data prediction on vehicle trajectory forecasting task, and propose a hybrid model that combines the merits of MLP and LSTM.
	Bike Sharing System Propose the first neural network-based framework integrated with fuzzy clustering for predicting per-station demand and status in bike-sharing systems, achieving state-of-the-art performance on the New York CitiBike dataset.
Undergraduate Thesis	Gradual Knowledge Distillation Video-based Human Action Detection. Undergraduate Degree Thesis, 2020 Gradually distill knowledge from higher-precision model to lower-precision and finally binary model to alleviate the performance deterioration in Quantization.

The website template was borrowed from source code.