Li MA 马力

I'm currently a Research Scientist at Scanline VFX Studio working with Paul Debevec. I obtained my PhD from the Dept. of Computer Science & Engineering in the Hong Kong University of Science and Technology (HKUST). My PhD advisor is Prof. Pedro V. SANDER, and I worked closely with Prof. Liao Jing. Before that, I received my Bachelor's Degree in EE from Zhejiang University in 2019.

I was a research intern at Meta Reality Labs from May to October 2023, with Michael Zollhoefer and Christian Richardt as my managers. Previously, I was a research intern at Tencent AI Lab under the guidance of Xiaoyu Li from August 2021 to September 2022.

I'm interested in Novel View Synthesis, Neural Rendering, Differentiable Rendering, and other graphics-related vision topics.

Before I delve into neural rendering research, I've also engaged in some low-level image and video processing tasks, including image/video stitching, inpainting, HDR creation, style transfer, video stabilization, and optical flow estimation.

Email  /  Github  /  Google Scholar  /  Twitter  /  LinkedIn  /  CV

profile photo
Publications

Representative papers are highlighted.

SpecNeRF: Gaussian Directional Encoding for Specular Reflections
Li Ma, Vasu Agrawal, Haithem Turki, Changil Kim, Chen Gao, Pedro V. Sander, Michael Zollhöfer, Christian Richardt
CVPR, 2024 (highlight)
project page / arXiv / pdf

We propose a Gaussian directional encoding that leads to better modeling of specular reflections under near-field lighting conditions.

3D Video Loops from Asynchronous Input
Li Ma, Xiaoyu Li, Jing Liao, Pedro V. Sander
CVPR, 2023
project page / live demo (VR demo) / Arxiv / pdf / code

We construct a novel 3D looping video representation that can do novel view synthesis in real-time on an iPhone X!

Apr. 2023 update: we add VR support to the live demo! (see the VR demo above)

Neural Parameterization for Dynamic Human Head Editing
Li Ma, Xiaoyu Li, Jing Liao, Xuan Wang, Qi Zhang, Jue Wang, Pedro V. Sander
Siggraph Asia, 2022
project page / arXiv / pdf / code

We try to introduce explicit parameters into implicit dynamic NeRF representations to achieve editing of 3D human heads.

Deblur-NeRF: Neural Radiance Fields from Blurry Images
Li Ma, Xiaoyu Li, Jing Liao, Qi Zhang, Xuan Wang, Jue Wang, Pedro V. Sander
CVPR, 2022
project page / arXiv / pdf / code

We optimize a NeRF and the blur kernels jointly from multi-view blurry images to get a sharp NeRF.

TVConv: Efficient Translation Variant Convolution
Jierun Chen, Tianlang He, Weipeng Zhuo, Li Ma, Sangtae Ha, S.-H. Gary Chan
CVPR, 2022
arXiv / code

We overparameterize a weight-generating block during training for efficient inference of dynamic convolutions.

AutoMatch: Leveraging Traffic Camera to Improve Perception and Localization of Autonomous Vehicles
Yuze He, Li Ma, Jiahe Cui, Zhenyu Yan, Guoliang Xing, Sen Wang, Qintao Hu, Chen Pan
SenSys, 2022
pdf / slides

We exploit existing ubiquitous traffic cameras to assist perception and localization in autonomous driving.

VI-Eye: Semantic-based 3D Point Cloud Registration for Infrastructure-assisted Autonomous Driving
Yuze He, Li Ma, Zhehao Jiang, Yi Tang, Guoliang Xing
MobiCom, 2021
pdf / fast forward / presentation / slides

We align two point clouds with little overlap using saliency points in the traffic scenario.

Other Projects

Some interesting projects that I worked on, including research projects.

3D Gaussian Splatting Viewer
Toy Project, 2023

code

A PyOpenGL based 3D Gaussian Splatting Viewer.

Dec. 2023 update: The viewer now supports CUDA backend! It's useful for debugging the official CUDA rasterizer.

The Replate
Li Ma, Ge Chen, Pedro V. Sander
Funded Project, 2021
project page / code

A Qt-based software that allows you to create infinitely looping videos from sports videos.

Layered Multiplane Videos for Novel View Synthesis
Li Ma, Jing Liao, Pedro V. Sander
Research Project, 2021
project page / pdf / supple / code

We try to generate a temporally consistent Multiplane Video (MPV) from only monocular video to achieve space and time control. The results are actually not as good as expected. But still, I post this project as a constant reminder for me to evaluate the feasibility before expending too much effort.

Boolean Text Geometry
Li Ma
Course Project, 2020
live demo / code

An interesting toy project based on Threejs that renders geometries that looks like different texts from different views.

Intelligent Car Racing
Kai Wang*, Li Ma*, Yuze He*
The NXP Cup, 2018
video (group round) / video (finals)

Exciting car racing!

Teaching Assistant

  • COMP5411 Advanced Computer Graphics, 2022-23 Fall
  • COMP5411 Advanced Computer Graphics, 2020-21 Fall
  • COMP2011 Programming with C++, 2019-20 Spring
  • Professional Activities

    Reviewer for Siggraph Asia 2024, CVPR 2024, TOG, ECCV 2024, Siggraph 2024, PR, TMM, TPAMI, WACV 2024, ICCV 2023, CVPR 2023, BMVC 2021.

    Selected Awards

  • Postgraduate Scholarship, HKUST
  • First-Class Scholarship for Outstanding Merits, ZJU
  • First-Prize of the NXP Cup National University Students Intelligent Car Race (Zhejiang Division), China
  • Second-Prize of the NXP Cup National University Students Intelligent Car Race, China
  • Excellent Student Awards, ZJU
  • Research and Innovation Scholarship, ZJU
  • Thanks to Jon Barron for sharing the source code of his personal webpage.