About:
I am a Principal Engineer at HUAWEI, working on 3D generative foundation models and world models. I own model R&D for the Xiaoyi 3D foundation model, which powers the team product Xiaoyi 3D × Pixel Artisan PRO. My latest work, SUPRA-Tex, turns a single image into a relightable PBR 3D asset whose textures stay aligned with the input; trained only on open-source data, it is competitive with leading commercial 3D generators. I work closely with Dr. Jiemin Fang and Dr. Qi Tian.
I received my Ph.D. in Computer Science and Technology from Shanghai Jiao Tong University (SJTU) in 2025, supervised by Prof. Wei Shen, after earning my B.S. (2019) and M.S. (2021) at SJTU.
In 2023, I received the MICCAI Young Scientist Award for LerPlane, my work on fast 4D reconstruction of deformable tissues.
I have published 20+ papers and preprints, with nearly 2,000 citations on Google Scholar.
Research Vision:
My long-term goal is to build world models for embodied intelligence: models that can generate, reconstruct, and simulate complex 3D environments. My current work follows three threads: 3D generative foundation models (UniLat3D, SUPRA-Tex, SCULPT), large-scale 3D world generation (WorldGrow, WorldAct), and high-quality 3D reconstruction (GaussianObject), which I aim to bring together into general-purpose spatial intelligence. Representative projects release code or complete project materials for reproducible evaluation and fair comparison; see my open research record. I believe world models will become a foundation of embodied AI and a critical step toward AGI.
2026-10:🚀 Our latest work SUPRA-Tex is now fully open source! From a single image, it builds a PBR 3D asset whose textures stay locked to the reference, down to fine lettering and patterns. Trained only on open-source data, it is preferred over five commercial systems (Hunyuan3D, Rodin, Tripo, Meshy, and Hi3D) for both visual quality and input alignment across 512 paired human evaluations. Project / Paper / Code / Model weights / Live demo.
2026-08:🚀SCULPT is released, introducing subtractive composition for 3D part generation.
2026-05:🌍WorldAct, a project I initiated and mentored, is released and now accepted to NeurIPS 2026. It turns monolithic 3D worlds into interaction-ready, object-centric scenes.
2025-06:📘 Our book Hands-on Computer Vision is published in English and Chinese (动手学计算机视觉). Written for computer vision practitioners and researchers, it can also serve as a textbook for computer vision courses in AI-related programs.
SUPRA-Tex: 3D Generation with Super-Aligned TextureChen Yang*, Jiemin Fang*, Fanpeng Meng*, Zihao Huang, Jichen Pang, Guanjun Wu, Zhikuan Bao, Qi Tian
First author and lead developer: Chen Yang arXiv, 2026 · Fully open source
From a single image, SUPRA-Tex builds a relightable PBR 3D asset whose textures stay locked to the reference, down to fine lettering, patterns, and material boundaries. 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. The 2.1B‑parameter model, code, and live demo are fully open.
