Butian Xiong
Butian Xiong
MSCS · USC Viterbi
Research Assistant · USC ICT
 CV  GitHub  Scholar  Email  Twitter

About Me

I am a Master's student in Computer Science at the University of Southern California (Viterbi School of Engineering) and a Research Assistant at USC Institute for Creative Technologies, supervised by Andrew Feng.

My research focuses on 3D scene understanding — including large-scale 3D Gaussian Splatting, semantic-aware 3D reconstruction, indoor navigation, and multimodal LiDAR–image fusion. I am passionate about building 3D perception systems that can reason about the real world at scale.

I am also currently interning at Futurewei Technologies (Huawei's US R&D arm), working on cutting-edge research in the industry. Previously, I worked as an AI Engineer at the Shenzhen Future Network Intelligence Institute and as a Research Scientist at Infused AI, collaborating with Yatong Han. I also served as a technical consultant at Auki-Labs. Earlier I was a Research Assistant at Prof. Zhen Li's DeepBit Lab (CUHK-SZ), the Geometry AI Lab (KAIST, supervised by Prof. Minhyuk Sung), the RAIL Lab (CUHK-SZ, supervised by Prof. Tin Lun Lam), and SRIBD (supervised by Prof. Liu Li).

I completed my B.S. in Computer Engineering with First Class Honors from the Chinese University of Hong Kong, Shenzhen in 2024, ranking 6th out of 103 in my major.

Education

2024 – 2026

M.S. in Computer Science — University of Southern California

Viterbi School of Engineering  ·  Los Angeles, CA

Spring 2023

Exchange — Korea Advanced Institute of Science & Technology (KAIST)

Department of Electrical Engineering  ·  Daejeon, South Korea

2020 – 2024

B.S. in Computer Engineering — Chinese University of Hong Kong (Shenzhen)

School of Science and Engineering  ·  First Class Honors  ·  Rank 6 / 103

Publications

NanoGS teaser
NanoGS: Training-Free Gaussian Splat Simplification
Butian Xiong*, Rong Liu*, Tiantian Zhou, Meida Chen, Zhiwen Fan, Andrew Feng (*equal contribution)
arXiv  Mar 2026
A training-free method for compressing 3D Gaussian Splat representations by formulating simplification as local pairwise merging over a sparse spatial graph. Uses mass-preserved moment matching to merge Gaussian pairs while preserving standard 3DGS parameterization — runs on CPU with no post-training optimization or calibrated images required.
arXiv Code Project
Splat Feature Solver poster
Splat Feature Solver
Butian Xiong, Rong Liu, Kenneth Xu, Meida Chen, Andrew Feng
ICLR 2026
We cast feature lifting on splat-based 3D representations as a sparse linear inverse problem with a closed-form solution, plus two regularizers (Tikhonov Guidance, Post-Lifting Aggregation) to stabilize and denoise multi-view features. Delivers state-of-the-art open-vocabulary 3D segmentation in minutes.
OpenReview arXiv Code Project Video
OOD Detection
Class Relevance Learning for Out-of-Distribution Detection
Liguang Zhou, Butian Xiong, et al.
ICASSP 2025
A class relevance learning (CRL) framework that harnesses inter-class relationships within the OOD pipeline, significantly outperforming prior max-logit methods on generic image classification (Near OOD and Far OOD) benchmarks.
Paper
SA-GS teaser
SA-GS: Semantic-Aware Gaussian Splatting for Large Scene Reconstruction with Geometry Constraint
Butian Xiong, et al.
arXiv  May 2024
SA-GS leverages SAM and DINO to generate semantic masks and introduces a geometric complexity measurement function as soft regularization, guiding each Gaussian Splat within specific semantic regions for fine-grained 3D geometry reconstruction.
arXiv
GauU-Scene V2 teaser
GauU-Scene V2: Assessing the Reliability of Image-Based Metrics with Expansive LiDAR Image Dataset Using 3DGS and NeRF
Butian Xiong, Nanjun Zheng, Zhen Li
arXiv  Apr 2024
Extends U-Scene to 6.5+ km² with the first LiDAR–image alignment method for a drone-based dataset. Reveals contradictions between image-based metrics (SSIM, LPIPS, PSNR) and geometric metrics (Chamfer distance) in 3DGS and NeRF evaluation.
arXiv
GauU-Scene teaser
GauU-Scene: A Scene Reconstruction Benchmark on Large Scale 3D Reconstruction Dataset Using Gaussian Splatting
Butian Xiong, et al.
arXiv  Jan 2024
Introduces U-Scene, a 1.5+ km² outdoor benchmark coupling RGB imagery with LiDAR ground truth collected via DJI Matrix 300 drone with Zenmuse L1, covering both urban and academic environments.
arXiv
Motion Capture
MoCapDT: Temporal Guided Motion Capture Solving with Diffusion Transfer
Butian Xiong
Journal of Electronics and Information Science  2023
A modified diffusion model architecture for reconstructing joint locations from noisy 4D marker data, using temporal guidance in latent space. Outperforms autoencoder-based baselines on CMU Synthesized and real-world NC-SOFT datasets.
Paper Local PDF

Experience

Research Intern  ·  Futurewei Technologies

2025 – Present  ·  (Huawei US R&D)
Industry research internship at Futurewei, Huawei's US research and development arm.

Research Assistant  ·  USC Institute for Creative Technologies

Spring 2024 – Present  ·  Los Angeles, CA  ·  Supervisor: Andrew Feng
3D scene understanding, indoor navigation, multiview-consistent semantic segmentation, embodiment dataset collection.

AI Engineer  ·  Shenzhen Future Network Intelligence Institute

2024  ·  Shenzhen, China
Large-scale indoor scene reconstruction; semantic embedding within Gaussian Splatting representations.

Research Scientist  ·  Infused AI

2024  ·  Remote  ·  Collaboration with Yatong Han
Co-led the 3D scene understanding research team with Dr. Xiao Yu.

Research Assistant  ·  CUHK-SZ DeepBit Lab

Fall 2023 – Spring 2024  ·  Supervisor: Prof. Zhen Li
3D LiDAR scanning, Gaussian Splatting, large-scale scene benchmark construction (GauU-Scene).

Research Assistant  ·  KAIST Geometry AI Lab

Summer 2023  ·  Daejeon, South Korea  ·  Supervisor: Prof. Minhyuk Sung
3D point cloud denoising using diffusion models and transformer architectures.

Research Assistant  ·  CUHK-SZ RAIL Lab

Fall 2022  ·  Supervisors: Prof. Tin Lun Lam & Dr. Liguang Zhou
Out-of-distribution detection, scene understanding, entropy-based algorithms for safe robotic deployment.

Research Assistant  ·  Shenzhen Research Institute of Big Data (SRIBD)

Summer 2022  ·  Supervisor: Prof. Liu Li
3D digital face and human generation with deep learning.

Technical Consultant  ·  Auki-Labs

2023
Spatial computing and AR/VR consulting.

Awards

Service

Research Group

I co-lead a 3D scene understanding research team at Infused AI and DeepBit Lab with Dr. Xiao Yu. If you are interested in joining our research, feel free to reach out via email.

Current research assistants: Jiayue Dai, Yunya Wang, Yuetong Chen, Yihan Fang

Hobbies