About Me
I’m Rong Liu, a Computer Science PhD student at University of Southern California.
Previously, I have worked at USC Institute for Creative Technologies, United Imaging Intelligence, and Netflix Eyeline Studios.
My long-term research goal is to answer how we enable machines to understand, reconstruct, generate, and interact with our everyday physical world through computer vision, graphics, and machine learning.
Current research interests: 3D/4D Reconstruction and Generation, Neural Representation and Rendering, and World Modeling.
I am currently seeking 2027 summer research internships.
Education
University of Southern California
August 2026 - May 2030 (Expected)Doctor of Philosophy in Computer Science
University of Southern California
August 2022 - May 2024Master of Science in Computer Science (Honors Merit)
GPA: 4.0/4.0Dalian University of Technology
September 2018 - June 2022Bachelor of Engineering in Computer Engineering
GPA: 88/100Technical Skills
Programming: Python, C++, CUDA, JavaScript/WebGL
Machine Learning: PyTorch, diffusion models, generative modeling, model compression, transfer learning, object detection
3D Vision & Graphics: 3D Gaussian Splatting, Neural Radiance Fields (NeRF), differentiable and neural rendering, real-time rendering, novel view synthesis, 3D/4D reconstruction, world modeling, monocular SLAM, camera geometry, relighting and appearance modeling, mesh extraction
Tools: Blender, COLMAP, synthetic data generation, CMake, Git, Linux
Experiences
Research Intern
May 2026 - August 2026Netflix Eyeline Studios
Supervisor: Dr. Li Ma, Dr. Ning Yu and Dr. Paul Debevec- Investigated 3D/4D splatting representations for world modeling of dynamic scenes in virtual-production workflows
- Built an automated Blender pipeline for large-scale synthetic data generation with controllable lighting and camera conditions
- Developed an image-relighting diffusion model conditioned on lighting cues to enable controllable scene illumination
Research Engineer
May 2024 - May 2026USC Institute for Creative Technologies
Supervisor: Prof. Yue Wang and Prof. Andrew Feng- Developed deformable Beta kernels with CUDA-accelerated real-time rendering, reaching state-of-the-art fidelity with 45% of the parameters and 1.5x faster rendering than 3DGS-MCMC, and were the first splatting method to surpass state-of-the-art NeRFs; published at SIGGRAPH 2025
- Co-developed an online dense monocular SLAM system integrating 3D Gaussian Splatting with dynamic depth and camera pose updates for real-time scene reconstruction; published at I3D 2025
- Formulated 2D-to-3D feature lifting as a sparse linear inverse problem, enabling efficient closed-form solutions for high-quality 3D semantic features; published at ICLR 2026
- Co-developed a training-free splat compression method that runs on CPU and cuts primitive count while preserving appearance and geometry, requiring no calibrated images or post-optimization; published at ECCV 2026
Research Intern
June 2025 - August 2025United Imaging Intelligence
Supervisor: Dr. Zhongpai Gao and Dr. Ziyan Wu- Generalized 3D Gaussian Splatting to N-dimensional anisotropic Beta kernels that unify spatial, angular, and temporal dependency modeling, outperforming prior methods on static, view-dependent, and dynamic benchmarks while rendering in real time via a CUDA implementation; published at ICLR 2026
- Investigated radiance-field representations for reconstructing and rendering medical data
Research Assistant
May 2023 - May 2024USC Institute for Creative Technologies
Supervisor: Prof. Andrew Feng- Developed AtomGS to improve radiance-field geometry and rendering fidelity through atomized Gaussian proliferation and edge-aware normal supervision; published at BMVC 2024
- Built scalable Neural Radiance Field pipelines for reconstructing and rendering large terrain datasets
- Investigated mesh extraction from NeRF and 3D Gaussian Splatting representations for downstream geometry processing
Research Assistant
July 2020 - July 2022Dalian University of Technology
Supervisor: Prof. Liangtian Wan- Reframed radar signal sorting as object detection over pulse-stream images, training and benchmarking YOLO-MobileNet, Faster R-CNN, and Cascade R-CNN detectors
- Developed a multi-source deep transfer-learning pipeline that pre-trains on UAV-swarm pulses collected across separate reconnaissance areas and fine-tunes on the target area, improving sorting accuracy over baselines when target-domain data is scarce; published in Information Fusion and granted as patent CN113030958B
Teaching Assistant
September 2021 - June 2022Dalian University of Technology
- Program Design Basics and C Programming
- Object-Oriented Method and C++ Program Design
Selected Publications
NanoGS: Training-Free Gaussian Splat Simplification
Butian Xiong*, Rong Liu*, Tiantian Zhou, Meida Chen, Zhiwen Fan, Andrew Feng
ECCV 2026
NanoGS is a training-free framework for Gaussian Splat simplification, which achieves substantial simplification ratios while preserving fidelity and geometry without post-optimization or calibrated images.
Website | Paper | Web App | Code (Web) | Code (Python)
Universal Beta Splatting
Rong Liu, Zhongpai Gao, Benjamin Planche, Meida Chen, Van Nguyen Nguyen, Meng Zheng, Anwesa Choudhuri, Terrence Chen, Yue Wang, Andrew Feng, Ziyan Wu
ICLR 2026
Universal Beta Splatting generalizes 3D Gaussian Splatting to N-dimensional anisotropic Beta kernels, enabling controllable dependency modeling across spatial, angular, and temporal dimensions within a single representation.
Splat Feature Solver
Butian Xiong, Rong Liu, Kenneth Xu, Meida Chen, Andrew Feng
ICLR 2026
A unified, kernel- and feature-agnostic framework that formulates feature lifting as a sparse linear inverse problem, enabling efficient closed-form solutions with high-quality 3D semantic features.
Deformable Beta Splatting
Rong Liu*, Dylan Sun*, Meida Chen, Yue Wang†, Andrew Feng†
ACM SIGGRAPH 2025
Deformable Beta Splatting introduces deformable Beta Kernels with adaptive frequency control for both geometry and color encoding, capturing complex geometries and lighting while only using 45% parameters and rendering 1.5x faster than 3DGS-MCMC.
SplatMAP: Online Dense Monocular SLAM with 3D Gaussian Splatting
Yue Hu, Rong Liu, Meida Chen, Peter Beerel, Andrew Feng
ACM I3D 2025
A real-time monocular SLAM system that fuses 3D Gaussian Splatting with SLAM’s dynamic depth and pose updates via SLAM-Informed Adaptive Densification and Geometry-Guided Optimization.
AtomGS: Atomizing Gaussian Splatting for High-Fidelity Radiance Field
Rong Liu, Rui Xu, Yue Hu, Meida Chen, Andrew Feng
BMVC 2024
AtomGS proposes an atomized proliferation of Gaussians and edge-aware normal loss to refine Gaussian splatting, boosting geometric precision and rendering fidelity in novel-view synthesis.
Website | Paper | Code | GS Monitor | Poster
UAV swarm based radar signal sorting via multi-source data fusion: A deep transfer learning framework
Liangtian Wan, Rong Liu, Lu Sun, Hansong Nie, Xianpeng Wang
Information Fusion. 78 (2022): 90-101
A UAV-swarm–enabled deep transfer learning framework that fuses radar pulses across time and space to accurately classify and sort complex radar signals.
Preprints
Universal Photorealistic Style Transfer: A Lightweight and Adaptive Approach
Rong Liu, Enyu Zhao, Zhiyuan Liu, Andrew Feng, Scott John Easley
arXiv:2309.10011
A lightweight style transfer network that learns instance-adaptive photorealistic transfer on-the-fly and scales effortlessly to high-resolution images and videos.
Awards
-
Computer Science Master’s Student Honors Merit - University of Southern California (USC), 2024
-
DUT Outstanding Undergraduate Thesis - Dalian University of Technology (DUT), 2022