|
Yixuan Yang Hi there! My name is Yixuan Yang (杨翊玄, pronounced 'Yee-shwen Yang'). I am currently a Ph.D. candidate at DCS, University of Warwick, UK, supervised by Prof. Victor Sanchez and Prof. Feng Zheng (SUSTech, China). I received my Master degree at CSE, The Chinese University of Hong Kong (CUHK) in 2020. Previously, I obtained my Bachelor's degree in the School of Computer Science in 2019 at Northwest University, China. Now, I am an intern at Shanghai AI lab, working on 3D scene generation and 3D reconstruction. Besides, I am also a photography enthusiast, I share some of my photography works in this page. Email / Github / Photo Gallery |
|
News2026.5: One paper about simulation-ready tabletop layout generation is accepted by ICML 2026! 2025.9: One paper about 3D scene generation with Multi-stage DPO is accepted by NeurIPS 2025! 2025.6: One paper about 3D scene generation with LLMs is accepted by IROS 2025! 2025.3: One paper about unpaired point cloud completion is accepted by IEEE TMM! 2024.11: One paper about video anomaly detection is accepted by IEEE TIP! 2024.11: I'm excited to join Shanghai AI Lab as a research intern, focusing on 3D scene generation and reconstruction! |
ResearchMy research interests include 3D computer vision, 3D generation and 3D perception. And now I am working on 3D scene generation with LLMs and 3D reconstruction for the transparent and reflective objects. |
|
Code-as-Room: Generating 3D Rooms from Top-Down View Images via Agentic Code Synthesis
Yixuan Yang*, Zhen Luo*, Wanshui Gan*, Jinkun Hao, Junru Lu, Jinghao Yan, Zhaoyang Lyu, Xudong Xu * Equal contribution paper / project page / code We generate 3D rooms from top-down view images via agentic code synthesis with LLM agents. |
|
STABLE: Simulation-Ready Tabletop Layout Generation via a Semantics-Physics Dual System
Zhen Luo*, Yixuan Yang*, Xudong Xu, Jinkun Hao, Zhaoyang Lyu, Feng Zheng, Jiangmiao Pang, Yanwei Fu * Equal contribution International Conference on Machine Learning (ICML 2026) paper / project page A semantics-physics dual system for generating simulation-ready tabletop object layouts. |
|
OptiScene: LLM-driven Indoor Scene Layout Generation via Scaled Human-aligned Data Synthesis and Multi-Stage Preference Optimization
Yixuan Yang*, Zhen Luo*, Tongsheng Ding*, Junru Lu, Jinyu Yang, Mingqi Gao Victor Sanchez, Feng Zheng * Equal contribution arxiv / project page We introduce a new large 3D synthetic indoor layout dataset 3D-SynthPlace and introduce LLM-based Optiscene with two-stage DPO to generate new 3D layout scene. |
|
LLplace: The 3D Indoor Scene Layout Generation and Editing via Large Language Model
Yixuan Yang, Junru Lu, Zixiang Zhao, Zhen Luo, Wanxi Dong, Victor Sanchez, Feng Zheng 2025 IEEE International Conference on Intelligent Robots and Systems (IROS 2025) paper We use LLMs to generate and edit the 3D indoor scene layout. |
|
RefComp: A Reference-guided Unified Framework for Unpaired Point Cloud Completion
Yixuan Yang, Jinyu Yang, Zixiang Zhao, Victor Sanchez, Feng Zheng IEEE Transactions on Multimedia (TMM) paper We apply the reference-guided generation to guide the unpaired point cloud completion. |
|
Advancing Video Anomaly Detection:
A Bi-Directional Hybrid Framework for Enhanced Single- and Multi-Task Approaches
Guodong Shen, Yuqi Ouyang, Junru Lu, Yixuan Yang, Victor Sanchez IEEE Transactions on Image Processing (TIP) paper We introduce a bi-directional hybrid framework to detect anomaly frames of the video. |