Fanpeng Meng | 孟繁鹏

I am an Engineer at Huawei Inc. and a core member of the Xiaoyi 3D Team, working under the leadership of Dr. Jiemin Fang and Dr. Qi Tian.

Prior to that, I received an M.Phil. in Computer & Information Engineering from The Chinese University of Hong Kong (Shenzhen), advised by Prof. Xiaoguang Han. Before that, I obtained a B.Eng. from Huazhong University of Science & Technology, and once worked with Prof. He Wang from Peking University.

Research Interests: My research focuses on 3D AIGC & World Models. I aspire to build high-quality, practical, and efficient 3D digital worlds towards the AGI era.

Email  /  Google Scholar  /  GitHub

Fanpeng Meng

News

Research

[*] equal contribution; [†] corresponding author(s);
Representative papers are highlighted.

SUPRA-Tex: 3D Generation with Super-Aligned Texture
Chen Yang*, Jiemin Fang*†, Fanpeng Meng*, Zihao Huang, Jichen Pang, Guanjun Wu, Zhikuan Bao, Qi Tian†
arXiv, 2026 · Fully open source

From a single image, generating relightable PBR 3D assets with reference-aligned textures that preserve fine lettering, patterns, and material details. Trained only on open-source data, it is preferred over five commercial systems, Hunyuan3D 3.1, Rodin 2.5, Tripo H3.1, Meshy 7.1, and Hi3D 3.0, for both visual quality and input alignment across 512 paired human evaluations.

Webpage / Paper / Code / Weights / Demo
Gate3DGen: A Pre-processing Model that Selectively Purifies Real-World Images for 3D Generation
Xiaolin Li*, Zihao Huang*, Fanpeng Meng, Chen Yang, Guanjun Wu, Kai Zhang†, Jiemin Fang†, Qi Tian†
arXiv, 2026
Webpage / Paper / Code

Reducing blur, occlusion, and other image degradations while preserving object identity and details, providing higher-quality inputs for 3D generation.
Four SUPRA-Tex examples with enlarged lettering and material details, input references, and a consistent dark background
Gate3DGen's original video explains the input-image problem and preprocessing workflow, followed by blurred lettering, handheld-object, and color-cast examples with original and purified inputs
ProxyUp bread slicing, curtain opening, falling books, and Newton's cradle, each with a synchronized input proxy in the upper-right corner
DreamDissector animation showing an astronaut and kangaroo disentangled into independently editable 3D objects
3D-aware navigation method overview: online 3D mapping, simultaneous exploration and identification, and local planning, followed by a navigation demonstration

Career

Education


This website template was borrower from Dr. Jon Barron.
Last updated on Oct. 23, 2026.