Chuan FANG
Ph.D. Candidate
Electronic and Computer Engineering
Hong Kong University of Science and Technology
Hong Kong SAR, China
Email: cfangac@connect.ust.hk / fang1457737815@gmail.com
👤 Biography
Hi theređź‘‹, I am Chuan Fang (in Chinese: ć–ąĺ·ť), a fourth-year Ph.D. candidate in Electronic and Computer Engineering at Hong Kong University of Science and Technology, supervised by Prof. Ping Tan.
My research is driven by a single question: how can we build AI that truly understands and inhabits the physical world? I pursue this through Multi-modal World Models — AI systems that 👀perceive, 🎨generate, and 🦾interact within physical environments — along four tightly connected directions: (1) Multi-modal LLMs for Spatial Perception, (2) Generative Models for Controllable Video/3D Scene Synthesis, (3) Geometry-grounded Reasoning for General-purpose Spatial Intelligence, and (4) Agentic Frameworks that Unify Understanding and Generation, where perception, reasoning, and content creation converge into a single closed loop. Weaving these directions together, my research ambition is to build general-purpose intelligent systems capable of understanding, generating, and interacting with complex, real-world physical environments.
Before HKUST, I spent three formative years (2019–2023) as a Senior Algorithm Engineer at Alibaba DAMO Academy’s XR-Lab, working on 3D vision reconstruction and multi-sensor calibration under Prof. Ping Tan. After that, I’ve spent wonderful time as a research intern at ManyCore Tech for building world models that perceive, understand, and generate physical 3D space, working with Dr. Zihan Zhou.
I welcome research discussions and collaborations — please don’t hesitate to reach out.
📌 I will graduate in Spring 2027 and am actively seeking full-time research opportunities. If you think my background could be a strong fit for your team, I’d love to hear from you.
📢 News
- [2026-07] One paper about Visual-Language-Navigation (GeoDream) was accepted to ACMMM 2026.
- [2026-06] One paper about structured indoor modeling with LLMs (SceneSpinner) was accepted to ECCV 2026.
- [2025-11] One paper about layout-guided 3D indoor scene generation (SpatialGen) was accepted to 3DV 2026.
- [2025-09] One paper about structured indoor modeling with LLMs (SpatialLM) was accepted to NeurIPS 2025.
- [2025-06] One paper about geometry surface reconstruction in Gaussian Splatting (RaDeGS) was accepted to TOG 2025.
- [2024-11] One paper about text-to-3D scene mesh generation (CtrlRoom) was accepted to 3DV 2025.
- [2024-02] One paper about panoramic total scene understanding (PanoContext-Former) was accepted to CVPR 2024.
- [2021-07] One paper about multi-sensor calibration was accepted to IROS 2021.
đź“‘ Selected Publications [Google Scholar]
*: equal contribution
â– SpatialCrafter: Single-Image World Modeling via Generative 3D Proxies
Chuan Fang, Lingteng Qiu, Yixun Liang, Rui Chen, Yuantong Bai, Zhaohua Zheng, Feipeng Tian, Zilong Dong, Zihan Zhou, Ping Tan
ACM SIGGRAPH Asia 2026, submitted
TL;DR: SpatialCrafter builds generative 3D proxies from a single image to enable consistent, controllable video generation.
Yu Zhong, Zihao Zhang, Rui Zhang, Lingdong Huang, Shuo Wang, Chuan Fang, Xishan Zhang, Jiaming Guo, Shaohui Peng, Di Huang, Yanyang Yan, Xing Hu
ACM International Conference on Multimedia (ACMMM), 2026,
[Project Page] GitHub -- stars
TL;DR: Dreaming the Physical World via a sequential panoramic generative model to facilitate vision-and-language navigation.
â– H-OmniStereo: Zero-Shot Omnidirectional Stereo Matching with Heading-Aligned Normal Priors
Chenxing Jiang, Zhe Tong, Pusen Gao, Peize Liu, Yang Xu, Chuan Fang, Ping Tan, Shaojie Shen
IEEE Robotics and Automation Letters, submitted
[Project Page] GitHub -- stars
TL;DR: H-OmniStereo performs zero-shot omnidirectional stereo matching guided by heading-aligned normal priors.
â– SceneSpinner: Taming LLMs for Codematic Indoor Scene Generation
Yixun Liang, Qianyi Wu, Chuan Fang, Rui Chen, Jiahang Liu, Jianfeng Zhang, and Ping Tan
European Conference on Computer Vision (ECCV), 2026,
[Project Page] GitHub -- stars
TL;DR: SceneSpinner tames LLMs for codematic indoor scene generation.
â– SpatialGen: Layout-guided 3D Indoor Scene Generation
Chuan Fang, Heng Li, Yixun Liang, Jia Zheng, Yongsen Mao, Yuan Liu, Rui Tang, Zihan Zhou, Ping Tan
International Conference on 3D Vision (3DV), 2026
[Project Page] GitHub -- stars
TL;DR: SpatialGen generates realistic, layout-consistent 3D indoor scenes conditioned on structured room layouts.
â– SpatialLM: Training Large Language Models for Structured Indoor Modeling
Yongsen Mao, Junhao Zhong, Chuan Fang, Jia Zheng, Rui Tang, Hao Zhu, Ping Tan, Zihan Zhou
Neural Information Processing Systems (NeurIPS), 2025
[Project Page] GitHub -- stars
TL;DR: SpatialLM trains large language models to directly predict structured 3D indoor scene representations.
â– RaDeGS: Rasterizing Depth in Gaussian Splatting
Baowen Zhang, Chuan Fang, Rakesh Shrestha, Yixun Liang, Xiaoxiao Long, Ping Tan
ACM Transactions on Graphics (TOG), 2025
[Project Page] GitHub -- stars
TL;DR: RaDeGS improves geometric fidelity of Gaussian Splatting by directly rasterizing depth during optimization.
â– CtrlRoom: Controllable Text-to-3D Room Meshes Generation with Layout Constraints
Chuan Fang, Yuan Dong, Kunming Luo, Xiaotao Hu, Rakesh Shrestha, Ping Tan
International Conference on 3D Vision (3DV), 2025
[Project Page] GitHub -- stars
TL;DR: CtrlRoom generates controllable, layout-constrained 3D room meshes directly from text descriptions.
â– PanoContext-Former: Panoramic Total Scene Understanding with a Transformer
Yuan Dong*, Chuan Fang*, Liefeng Bo, Zilong Dong, Ping Tan
Computer Vision and Pattern Recognition (CVPR), 2024
[Project Page] GitHub -- stars
TL;DR: PanoContext-Former jointly reasons about layout, objects, and semantics in panoramic scenes with a transformer.
â– Single-Shot is Enough: Panoramic Infrastructure Based Calibration of Multiple Cameras and 3D LiDARs
Chuan Fang, Shuai Ding, Zilong Dong, Honghua Li, Siyu Zhu, Ping Tan
International Conference on Intelligent Robots and Systems (IROS), 2021
TL;DR: A single-shot calibration method for multiple cameras and 3D LiDARs using panoramic infrastructure cues.
- â– RenderNet: Visual Relocalization Using Virtual Viewpoints in Large-Scale Indoor Environments
Rui Huang, Chuan Fang, Kejie Qiu, Le Cui, Zilong Dong, Siyu Zhu, Ping Tan, arXiv 2022. - â– AR Mapping: Accurate and Efficient Mapping for Augmented Reality
Jiahui Zhang, Shitao Tang, Kejie Qiu, Rui Huang, Chuan Fang, Le Cui, Zilong Dong, Siyu Zhu, Ping Tan, arXiv 2021.
🎓 Educations
- Ph.D. student, Electronic and Computer Engineering, Hong Kong University of Science and Technology 2023.02 – now
- M.Sc., Electronic Science and Engineering School, Nanjing University 2016.09 – 2019.06
- B.E., Automation School, Nanjing University of Posts and Telecommunications 2012.09 – 2016.06
đź“ť Services & Talks
- â– Conference Reviewer:
- Computer Vision: ICCV 2024–2025, CVPR 2024–2026, ECCV 2025–2026
- Graphics: SIGGRAPH Asia 2026, Eurographics 2025, 3DV 2025–2026
- Robotics: IROS 2026
- â– Journal Reviewer: TVCG
- â– Invited talk: Indoor Scene Understanding and Generation, AI Center, Beike Inc., 2024.01. [[Slides]]
- â– Invited talk: Multiple Sensor Calibration, 3D Vision Online WeChat Account, 2021.06. [[Video]]
đź’» Experiences
- Research Intern, ManyCore Tech, China 2024.06 – now
- LightIllusion, China 2023.06 – 2024.06
- Senior Algorithm Engineer, DAMO Academy XR-Lab, Alibaba Group 2019.06 – 2023.01