Liyuan Zhu

I am a PhD student at Stanford University, advised by Prof. Iro Armeni and Prof. Gordon Wetzstein. I am a research scientist intern at NVIDIA Research. I obtained my master's at ETH Zurich, working with Prof. Konrad Schindler and Dr. Shengyu Huang. Before this, I obtained my bachelor's degree from Wuhan University.
My research goal is reconstructing and generating interactive 4D world. I am interested in general computer vision and graphics problems.
Office: Y2E2, 473 Via Ortega, Stanford, 94305 CA
Email / Scholar / Github / LinkedIn / Twitter

Publications/Preprints

* denotes equal contribution

GaussFusion: Improving 3D Reconstruction in the wild with a Geometry-informed Video Generator

Liyuan Zhu, Manjunath Narayana, Michal Stary, Will Hutchcroft, Gordon Wezstein, Iro Armeni
In CVPR, 2026
(Coming soon)

An efficient and generalizable 3D fixer using video generation.

WildPose: A Unified Framework for Robust Pose Estimation in the Wild

Jianhao Zheng, Liyuan Zhu, Zihan Zhu, Iro Armeni
In CVPR, 2026
(Coming soon)

A robust camera pose estimator for dynamic scenes.

Yue Pan, Tao Sun, Liyuan Zhu, Lucas Nunes, Iro Armeni, Jens Behley, Cyrill Stachniss
In arxiv preprint, 2026
Code / Project / Preprint

Towards a foundational 3D registratration model.

ReStyle3D: Scene-level Appearance Transfer with Semantic Correspondences

Liyuan Zhu, Shengqu Cai*, Shengyu Huang*, Gordon Wezstein, Naji Khosravan, Iro Armeni
In ACM SIGGRAPH technical papers, 2025
Code / Project / Preprint

Compositional 3D scene stylization with diffusion models.

Rectified Point Flow: Generic Point Cloud Pose Estimation

Tao Sun*, Liyuan Zhu*, Shengyu Huang, Shuran Song, Iro Armeni (* equal contribution)
In NeurIPS, 2025 (Spotlight)
Code / Project / Preprint

A generative formulation for point cloud pose estimation.

LoopSplat: Loop Closure by Registering 3D Gaussian Splats

Liyuan Zhu, Yue Li, Erik Sandström, Shengyu Huang, Konrad Schindler, Iro Armeni
In International Conference on 3D Vision (3DV), 2025 (Oral)
Code / Project / Preprint

State-of-the-art loop closure solution for RGB-D SLAM systems.

Living Scenes: Multi-object Relocalization and Reconstruction in Changing 3D Environments

Liyuan Zhu, Shengyu Huang, Konrad Schindler, Iro Armeni
In CVPR, 2024 (Highlight)
Code / Project / Paper

Understanding multi-object rigid motion over time. (Multi-body registration).

DeFlow: Self-supervised 3D Motion Estimation of Debris Flow

Liyuan Zhu, Yuru Jia, Shengyu Huang, Nicholas Meyer, Andreas Wieser, Konrad Schindler, Jordan Aaron
In CVPR Workshop, 2023 (Best paper award)
Code / Project / Paper

A self-supervised method to estimate 3D motion at pixel level.

Honors & Awards

NVIDIA Academic Grant Program Award on Video World Models, 2026.
Outstanding Reviewer (top 5%), CVPR 2025.
SPIRE Stanford Student Impact Fund Grant - Interior Design Using GenAI Models. 2025.
Culmann Prize (Best Master's Thesis), D-BAUG, ETH Zurich, Nov. 2023.
Best paper award, CVPR Workshop in Photogrammetric Computer Vision, 2023.

Talks

[ReStyle3D] SIGGRAPH 2025 oral presentation, in Vancouver, Aug. 12, 2025.
[ReStyle3D] Invited talk at NVIDIA Graphics Forum, July 8, 2025.
[LoopSplat] Oral presentation at 3DV 2025 in Singapore, Mar. 27, 2025,.
[LoopSplat] Invited talk at AnySyn3D (https://anysyn3d.github.io/) on Sep 19, 2024.

Experiences

Head TA, "Computer Vision for the Built Environment" CEE 247C, Stanford University, Winter Quarter, 2024 & 2025.
Teaching Assistant, "Multivariate Statistics and Machine Learning," Spring semester, ETH Zurich, 2023 fall.
Research Assistant, Photogrammetry and Remote Sensing, Konrad Schindler, ETH Zurich, Feb. 2022 - Jun. 2023.
Research Assistant, Hong Kong University of Science and Technology, Jun. 2020 - Sep. 2020

Academic Services

Reviewer: ICCV '23&25, CVPR '24&25&26, ECCV '24, NeurIPS '25

Page updated

Google Sites

Report abuse