Yu Liu @ THU

Yu Liu (刘宇)

I'm Yu Liu, a first year Ph.D. student in Department of Automation, Tsinghua University, advised by Prof. Song-Chun Zhu. I obtained my bachelor's degree in engineering from the Department of Automation at Tsinghua University, as the monitor of the Tong Class (an AGI program founded by Prof. Song-Chun Zhu). I'm currently working at General Vision Lab in BIGAI (Beijing Institute for General Artificial Intelligence) as a research intern advised by Dr. Baoxiong Jia and Dr. Siyuan Huang.

My research interest lies in computer vision, specifically unsupervised object-centric learning, articulated object reconstruction and 3D/4D reconstruction/generation. My hobbies are reading, music, natural scenery and meditation.

Email / Google Scholar / Github

Research

	Building Interactable Replicas of Complex Articulated Objects via Gaussian Splatting Yu Liu, Baoxiong Jia, Ruijie Lu, JunFeng Ni, Song-Chun Zhu, Siyuan Huang ICLR 2025 [Paper] [Project Page] [Code] We introduce ArtGS, a novel approach that leverages 3D Gaussians to reconstruct articulated objects from 2 states of RGBD images, which achieves state-of-the-art performance in joint parameter estimation and part mesh reconstruction.
	Decompositional Neural Scene Reconstruction with Generative Diffusion Prior Junfeng Ni, Yu Liu, Ruijie Lu, Zirui Zhou, Song-Chun Zhu, Yixin Chen, Siyuan Huang CVPR 2025 [Paper] [Project Page] [Code] We introduce DP-Recon, a novel approach that distills off-the-shelf diffusion models for high-quality decompositional scene reconstruction from sparse view images. Further fine-grained editings are well supported.
	TACO: Taming Diffusion for in-the-wild Video Amodal Completion Ruijie Lu, Yixin Chen, Yu Liu, Jiaxiang Tang, Junfeng Ni, Diwen Wan, Gang Zeng, Siyuan Huang ICCV 2025 [Paper] [Code] [Data] [Project Page] We introduce TACO, which repurposes pre-trained video diffusion models for Video Amodal Completion (VAC), facilitating downstream tasks like reconstruction. The key insight lies in the data curation and progressive training strategy.
	MOVIS: Enhancing Multi-Object Novel View Synthesis for Indoor Scenes Ruijie Lu, Yixin Chen, Junfeng Ni, Baoxiong Jia, Yu Liu, Diwen Wan, Gang Zeng, Siyuan Huang CVPR 2025 [Paper] [Code] [Data] [Project Page] We introduce MOVIS, which repurposes pre-trained diffusion models for multi-object level novel view synthesis (NVS) in indoor scenes. The key insight lies in incorporating a structure-aware noise scheduler and an auxiliary mask prediction task under novel views.
	SlotLifter: Slot-guided Feature Lifting for Learning Object-centric Radiance Fields Yu Liu, Baoxiong Jia, Yixin Chen, Siyuan Huang ECCV 2024 [Paper] [Project Page] [Code] We propose SlotLifter, a novel object-centric radiance model that aims to address the challenges of scene reconstruction and decomposition via slot-guided feature lifting.
	Improving Object-centric Learning With Query Optimization Baoxiong Jia, Yu Liu, Siyuan Huang ICLR 2023 [Paper] [Project Page] [Code] We proposed BO-QSA for (1) initializing Slot-Attention modules with learnable queries and (2) optimizing the model with bi-level optimization.

Experience

	Tsinghua University, China 2024.08 - now Ph.D. Student Advisor: Prof. Song-Chun Zhu
	Tsinghua University, China 2020.08 - 2024.07 Undergraduate Student Advisor: Prof. Song-Chun Zhu
	Beijing Institute for General Artificial Intelligence(BIGAI), China 2021.09 - now Research Intern Advisor: Dr. Baoxiong Jia and Dr. Siyuan Huang

Selected Awards and Honors

Tsinghua University outstanding undergraduate thesis
Student of the Year 2023, Department of Automation, Tsinghua University
Tsinghua University Comprehensive Excellent Award (2020 & 2022)
Tsinghua Science and Technology Innovation Excellence Award (2021)
Research Star Award, Beijing Institute for General Artificial Intelligence(BIGAI)
Huang Yicong Scholarship, Research Excellence Award

Thanks for your visiting by 😊.
Template stolen from Jon Barron. Page icon made by Freepik from flaticon.com .