Yiheng Zhang 张一恒

I'm now a research scientist intern at Tencent Games AI Center, collaborating with Professor Cheng Lin and Le Wan. Meanwhile, I am working closely with Professor Yuan Liu and Professor Wenping Wang.

Prviously, I received my M. Eng. degree from National University of Singapore. I also worked as an research intern at Tsinghua University with Researcher Yuwang Wang and Academian Qionghai Dai. I am truly fortunate that my interest in computer graphics was sparked and nurtured under the guidance of Professor Mirela Ben-Chen at the Center for Graphics and Geometric Computing in Technion - Israel Institute of Technology.

Email  /  CV  /  Linkedin  /  Github

profile photo

News

[Jan 2026] PartSAM has been accepted to ICLR 2026!
[Dec 2025] Attended SIGGRAPH Asia 2025 in Hong Kong, China.

Research Experience

🎓 I am actively seeking PhD opportunities for Fall 2026. Please feel free to reach out if you think I might be a good fit.

Research

My research interest lies in:

  • 3D Vision: Object & Scene-level Generation,
  • Computer Graphics: Geometry Processing and Geometric Learning (Operators & Quad),
  • Agent: Agentic Game/Movie Understanding & Design,
  • Robotics: Physics-based & Cognitive/Psychological Simulation.

💡 I am always open to discussions and collaborations, feel free to contact me anytime!

QuadLink: Autoregressive Quad-Dominant Mesh Generation via Point-Relation Learning
Under Review

Yiheng Zhang, Zhe Zhu, Tingrui Shen, Zhuojiang Cai, Tianxiao Li, Zixing Zhao, Qiujie Dong, Zhiyang Dou, Jiepeng Wang, Le Wan, Yuwang Wang, Wengping Wang, Yuan Liu†, Cheng Lin†
under review, 2026

M3DLayout: A Multi-Source Dataset of 3D Indoor Layouts and Structured Descriptions for 3D Generation
Under Review

Yiheng Zhang*, Zhuojiang Cai*, Mingdao Wang*, Meitong Guo, Tianxiao Li, Li Lin, Yuwang Wang†
under review, 2026
project page / arXiv / Github

FlashMesh: Faster and Better Autoregressive Mesh Synthesis via Structured Speculation
Under Review

Tingrui Shen*, Yiheng Zhang*, Chen Tang, Chuan Ping, Zixing Zhao, Le Wan, Yuwang Wang, Ronggang Wang, Shengfeng He†
under review, 2026
project page / arXiv / Github

PartSAM: A Scalable Promptable Part Segmentation Model Trained on Native 3D Data
ICLR 2026

Zhe Zhu, Le Wan, Rui Xu, Yiheng Zhang, Honghua Chen, Zhiyang Dou, Cheng Lin, Yuan Liu†, Mingqiang Wei†
International Conference on Learning Representations (ICLR), 2026
project page / arXiv / Github

LSS3D: Learnable Spatial Shifting for Consistent and High-Quality 3D Generation from Single-Image
IEEE MMSP 2025

Zhuojiang Cai*, Yiheng Zhang*, Meitong Guo, Mingdao Wang, Yuwang Wang†
accepted by IEEE MMSP, 2025 (IF: 7.3)
arXiv

Animation image transfer using CycleGAN
ICCAID 2022

Zhixun Liu*, Yiheng Zhang*, Xinyao Han, Wanting Zhou
accepted by ICCAID, 2022

Project
Tri2Quad: Geometry-Aware Triangle-to-Quad Mesh Conversion Operator
Advisor: Professor Cheng Lin, Professor Yuan Liu, Professor Wenping Wang

Github

I Developed a high-performance triangle-to-quad dominant conversion operator, formulated as a maximum-weight matching problem on the triangle adjacency graph and solved with modified Blossom matchers for quad-dominant remeshing, which integrated rich geometric filters into candidate edge generation before optimization to improve solver efficiency, with globally consistent normals' ensurance. The operator has been the key data-curation step of my HybridGen SIGGRAPH paper and deployed in Tencent's production pipeline, achieving state-of-the-art quad quality compared with PyMeshLab, Blender's built-in remesher, and Hunyuan's ILP-based operator.

Hybrid Intrinsic–Extrinsic Spectral Operators for DiffusionNet
Advisor: Professor Mirela Ben-Chen

Github

Replaced and complemented the standard Laplace-Beltrami operator in DiffusionNet with Steklov and elastic-basis operators, enabling the network to jointly encode intrinsic surface geometry and extrinsic embedding information. Demonstrated consistent improvements over the original DiffusionNet baseline on mesh segmentation, correspondence, and feature extraction benchmarks, showcasing a concrete case where classical geometry-processing operators and deep learning architectures are tightly integrated for better geometric understanding.


This page is adapted from this template.