Chenguo Lin

👤 Biography

Hi there👋, my name is Chenguo Lin (in Chinese: 林琛果). I am a final-year Ph.D. candidate of computer science and artificial intelligence at Peking University, China, supervised by Prof. Yadong Mu.

My research focuses on Multi-modal World Models: intelligent systems that can 👀 perceive, 🧠 reason, and 🦾 act across both the physical world and the digital world. I view world models as the substrate for recursive self-improvement (RSI): they turn interaction into verifiable experience, enabling agents to continuously expand both their understanding of the world and their ability to act within it. I am particularly interested in: (1) 🌍Physical World Models: controllable video/3D/action generation with spatial intelligence, for understanding, simulating, and interacting with the visual/physical environment. (2) 💻Digital World Models: GUI agents and computer-using agents (CUA) that perceive screens, reason over interfaces, and interact through tools, extending world modeling to dynamic and interactive digital environments.

I've spent wonderful time as a research intern at ByteDance Seedance for multi-modal video generation, ByteDance PICO for 3D generation and spatial intelligence, Microsoft Research Asia (MSRA) for self-supervised learning with Dr. Zhirong Wu and Dr. Stephen Lin, and as a research assistant at HKU with Prof. Ping Luo, and KAIST with Dr. Chaoning Zhang and Prof. In So Kweon.

I'm always open to research discussions and collaborations. Feel free to 📲contact me if you are interested.

📢 News

• [2026-09] Two papers for video world modeling (ReGen & JAG) were accepted to NeurIPS 2026.
• [2026-06] One paper about long-term memory for video generation (Ring Forcing) was accepted to ECCV 2026.
• [2026-05] Awarded Presidential Scholarship of Peking University.
• [2026-04] Awarded Meshy Fellowship.
• [2026-02] Three papers about 4D/video generation (MoVieS & Diff4Splat & ID-Crafter) were accepted to CVPR 2026.
• [2025-12] Awarded National Scholarship of China.
• [2025-09] Two papers about 3D/4D content creation (PartCrafter & DynamicVerse) were accepted to NeurIPS 2025.
• [2025-08] A unified 2D & 3D instruction-to-layout generation framework (InstructLayout) was accepted to T-PAMI 2025.
• [2025-07] We released a feed-forward model for 4D content creation from monocular videos: MoVieS .
• [2025-06] We released a 3D-native DiT that generates 3D objects in parts: PartCrafter .
• [2025-01] Two papers about 3D object & dynamics generation (DiffSplat & OmniPhysGS) were accepted to ICLR 2025.
• [2024-09] One paper about generalizable single-view human reconstruction (HumanSplat) was accepted to NeurIPS 2024.
• [2024-07] One paper about large-scale time-series pretraining (NuTime) was accepted to TMLR 2024.
• [2024-01] One paper about 3D scene synthesis (InstructScene) was accepted to ICLR 2024 as a spotlight paper.

📑 Selected Publications [Google Scholar]

*: equal contribution; †: project lead

NeurIPS 2026	ReGen: Agentic Video World Modeling with Synergized Reasoning and Generation Chenguo Lin, Yu Tang, Weiqiao Zheng, Enhua Jiang, Zhiguang Liu, Maomao Li,Muxi Chen, Xiaojia Chen, Shisheng Huang, Jianhuan Zhuo, Zilin Yang, Jiarong Ou, Rui Chen, Yadong Mu Neural Information Processing Systems (NeurIPS), 2026 [arXiv] [Project Page] [Code] TL;DR: ReGen is an agentic framework for long-horizon video world modeling that synergizes reasoning and generation with an action-specification interface.
arXiv 2026	4DStreamCtrl: Interactive Video Generation with Online 4D Control Shiqian Li, Chenguo Lin, Zhiguang Liu, Yu Tang, Jiarong Ou, Rui Chen, Yixin Zhu Preprint, arXiv [arXiv] [Project Page] [Code] TL;DR: 4DStreamCtrl is a real-time video generative model that allows users interactions in 4D space via 3D point tracking.
arXiv 2026	Guided End-to-End AutoRegression for Image Synthesis Bin Lin, Zheyuan Liu, Chenguo Lin, Sixiang Chen, Yunyang Ge, Yunlong Lin, Jianwei Zhang, Miles Yang, Zhao Zhong, Liefeng Bo, Li Yuan Preprint, arXiv [arXiv] [Project Page] [Code] TL;DR: GEAR co-trains a VQ tokenizer and an autoregressive (AR) image generator end-to-end, allowing much easier and faster training than existing methods.
ECCV 2026	Ring Forcing: Towards Precise Long-Term Memory for Autoregressive Video Diffusion Bowen Xue, Brandon Y. Feng, Chenguo Lin, Yuchen Lin, Yujia Zeng, Lvmin Zhang, Maneesh Agrawala, Honglei Yan, Panwang Pan European Conference on Computer Vision (ECCV), 2026 [arXiv] [Project Page] [Code] TL;DR: Ring Forcing is an autoregressive video diffusion framework designed to robustly construct and precisely utilize long-term memory.
NeurIPS 2026	Spatially-Grounded Long Video Generation with Self Geometry Forcing Chenguo Lin, Panwang Pan, Bowen Xue, Ruijie Lu, Yuchen Lin, Chenxin Li, Brandon Y. Feng, Yadong Mu Neural Information Processing Systems (NeurIPS), 2026 [arXiv] [Project Page] [Code] TL;DR: JAG is a spatially-grounded causal video diffusion model that integrates 3D perception for long and consistent video generation.
CVPR 2026	MoVieS: Motion-Aware 4D Dynamic View Synthesis in One Second Chenguo Lin*, Yuchen Lin*, Panwang Pan†, Yifan Yu, Tao Hu, Honglei Yan, Katerina Fragkiadaki, Yadong Mu Computer Vision and Pattern Recognition (CVPR), 2026 [arXiv] [Project Page] [Code] TL;DR: MoVieS is a feed-forward framework that jointly reconstructs appearance, geometry and motion for 4D scene perception from monocular videos.
CVPR 2026	Diff4Splat: Generalizable 4D Scene Generation with Latent Dynamic Reconstruction Models Panwang Pan*, Chenguo Lin*, Jingjing Zhao, Chenxin Li, Yuchen Lin, Haopeng Li, Honglei Yan, Kairun Wen, Yunlong Lin, Yixuan Yuan, Yadong Mu Computer Vision and Pattern Recognition (CVPR), 2026 [arXiv] [Project Page] [Code] TL;DR: Diff4Splat is a generalizable framework for controllable 4D scene generation from a single image using a video diffusion model.
CVPR 2026	ID-Crafter: VLM-Grounded Online RL for Compositional Multi-Subject Video Generation Panwang Pan, Jingjing Zhao, Yuchen Lin, Chenguo Lin, Chenxin Li, Hengyu Liu, Tingting Shen, Yadong Mu Computer Vision and Pattern Recognition (CVPR), 2026 [arXiv] [Project Page] [Code] TL;DR: ID-Crafter generates multi-subject videos combining text and reference images through identity-preserving RL and VLM-guided semantic alignment.
NeurIPS 2025	PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers Yuchen Lin*, Chenguo Lin*, Panwang Pan†, Honglei Yan, Yiqiang Feng, Yadong Mu, Katerina Fragkiadaki Neural Information Processing Systems (NeurIPS), 2025 Received more than 1.5K GitHub stars within a month [OpenReview] [arXiv] [Project Page] [Code] TL;DR: PartCrafter is a structured 3D generative model that jointly generates multiple parts and objects from a single RGB image in one shot.
NeurIPS 2025	DynamicVerse: Physically-Aware Multimodal Modeling for Dynamic 4D Worlds Kairun Wen*, Yuzhi Huang*, Runyu Chen, Hui Zheng, Yunlong Lin, Panwang Pan, Chenxin Li, Wenyan Cong, Jian Zhang, Junbin Lu, Chenguo Lin, Dilin Wang, Zhicheng Yan, Hongyu Xu, Justin Theiss, Yue Huang, Xinghao Ding, Rakesh Ranjan, Zhiwen Fan Neural Information Processing Systems (NeurIPS), 2025 [OpenReview] [arXiv] [Project Page] [Code] TL;DR: DynamicVerse is a large-scale multi-modal framework that integrates foundation models to convert videos into 4D representations, such as geometry, motion, semantics, etc.
ICLR 2025	DiffSplat: Repurposing Image Diffusion Models for Scalable Gaussian Splat Generation Chenguo Lin, Panwang Pan†, Bangbang Yang, Zeming Li, Yadong Mu International Conference on Learning Representations (ICLR), 2025 [OpenReview] [arXiv] [Project Page] [Code] TL;DR: DiffSplat directly generates 3D Gaussians by taming large text-to-image diffusion models from text prompts and single-view images in 1~2 seconds.
ICLR 2025	OmniPhysGS: 3D Constitutive Gaussians for General Physics-Based Dynamics Generation Yuchen Lin, Chenguo Lin†, Jianjin Xu, Yadong Mu International Conference on Learning Representations (ICLR), 2025 [OpenReview] [arXiv] [Project Page] [Code] TL;DR: OmniPhysGS synthesizes general physics-based 3D dynamic scenes, and can automatically and flexibly model various materials with domain-expert constitutive models.
NeurIPS 2024	HumanSplat: Generalizable Single-Image Human Gaussian Splatting with Structure Priors Panwang Pan*, Zhuo Su*†, Chenguo Lin*, Zhen Fan, Yongjie Zhang, Zeming Li, Tingting Shen, Yadong Mu, Yebin Liu Neural Information Processing Systems (NeurIPS), 2024 [OpenReview] [arXiv] [Project Page] [Code] TL;DR: HumanSplat digitalizes any human from a single input image in seconds via multi-view image diffusion models and latent Gaussian splatting reconstruction.
T-PAMI 2025	InstructLayout: Instruction-Driven 2D and 3D Layout Synthesis with Semantic Graph Prior Chenguo Lin*, Yuchen Lin*, Panwang Pan, Xuanyang Zhang, Yadong Mu Transactions on Pattern Analysis and Machine Intelligence (T-PAMI) [IEEE Xplore] [arXiv] [Project Page] [Code] TL;DR: InstructLayout is an extension of InstructScene that improves controllability and fidelity for the layout synthesis of both 2D E-commerce posters and 3D indoor scenes.
ICLR 2024	InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior Chenguo Lin, Yadong Mu International Conference on Learning Representations (ICLR), 2024 Spotlight (acceptance rate: 5%) [OpenReview] [arXiv] [Project Page] [Code] TL;DR: InstructScene is a generative framework to synthesize 3D indoor scenes from textual instructions, and is composed of a semantic graph prior and a layout decoder.
TMLR 2024	NuTime: Numerically Multi-Scaled Embedding for Large-Scale Time-Series Pretraining Chenguo Lin*, Xumeng Wen*, Wei Cao, Congrui Huang, Jiang Bian, Stephen Lin, Zhirong Wu† Transactions on Machine Learning Research (TMLR), 2024 [OpenReview] [arXiv] [Code] TL;DR: NuTime is a Transformer-based architecture that can take raw values of time-series data as input without any data normalization and transformation.

🎓 Educations

• Ph.D. student, School of Intelligence Science and Technology, Peking University
  2022 - 2027 (expected)
  
  
• B.Eng., College of Computer Science, Sichuan University
  2018 - 2022
  
  
  • Comprehensive Ranking: 1/171
  • Member of Honor College (for top 2% undergraduates at Sichuan University)

🏆 Honors & Awards

• Presidential Scholarship, Peking University (the highest honor scholarship in PKU)
  2026
• Meshy Fellowship, Meshy AI (only 10 winners globally per year)
  2026
• National Scholarship, Ministry of Education (the highest honor scholarship in China)
  2025
• Outstanding Student, Peking University
  2026, 2025
• ICLR 2024 Scholar Award, ICLR Committee
  2024
• Fresh Ph.D. Student Scholarship of WICT, Peking University (only 4 winners at WICT per year)
  2022
• Outstanding Bachelor Thesis, Sichuan University
  2022
• BaoSteel Scholarship, BaoSteel Education Foundation (only 6 winners in SCU per year, including postgraduates)
  2021
• National Scholarship, Ministry of Education (the highest honor scholarship in China)
  2021
• Outstanding Graduate, Sichuan University
  2021
• Outstanding Student (Cadre), Sichuan University
  2021, 2020, 2019
• Comprehensive Scholarship, Sichuan University
  2021, 2020, 2019

📝 Services

• Conference Reviewer:
  • Machine Learning: NeurIPS 2023-2026, ICLR 2024-2027, ICML 2024-2026
  • Computer Vision: ICCV 2023-2025, CVPR 2025-2026, ECCV 2026
  • Graphics: SIGGRAPH Asia 2025-2026
• Journal Reviewer: TMLR, TPAMI, TMM