Caoyuan Ma (马草原)

I am a master's student at the School of Computer Science of Wuhan University from Sep. 2022. my advisor is Professor Zheng Wang. From July 2022 to September 2023, I was fortunate to work at JD Explore as a research intern, with mentors Wu Liu and Xincheng Liu. Throughout my studies, I have been fortunate to collaborate closely with Zhixiang Wang, Xianzeng Ma, and Lixiong Chen.

News

  • 06/2024  I attended CVPR2024 in Seattle and I was invited to attend the Meshcapade dinner at June 19th by Yu Sun, it was a great week.
  • 05/2024  Serve as an MM24 reviewer.
  • 04/2024  I'm preparing for the ILETS test. I hope to get a 2025 fall PhD position in Singapore or Japan. Feel free to contact me if you have any chances or suggestions.
  • 04/2024  My new research TAAT: Think and Act from Arbitrary Texts already has a preliminary version, which can generate motions with texts processed by LLMs.
  • 03/2024  My first work HumanNeRF-SE was accepted by CVPR2024.
  • 06/2022  Outstanding graduates of Wuhan University.
  • 10/2021  National scholarship 国家奖学金 (the seconed time). (Award Rate: 0.2% nation-wide) Ministry of Education, China.
  • 10/2021  First Class Scholarship (Award Rate: 5% school-wide) Wuhan University.
  • 10/2020  National scholarship 国家奖学金. (Award Rate: 0.2% nation-wide) Ministry of Education, China.
  • 10/2020  First Class Scholarship (Award Rate: 5% school-wide) Wuhan University.

Research interests

Human-centric AIGC
  • Generative Computational Photography: leveraging generative models for image/video reconstruction and re-rendering.
  • Embodied AI: Not only possess advanced cognitive abilities such as perception, reasoning, and decision-making, but also have physical bodies and the ability to interact with the environment. To create a robot like ATRI is my dream.
  • AGI: A agent could perform any intellectual task that a human can, across a wide range of domains, without being explicitly programmed for each specific task. I don't like the current LLMs because they are being taught a lot of useless knowledge.
  • I believe the ideal AGI model only needs basic understanding and execution capabilities, and should be able to use tools like humans when completing specific tasks. The only cloud hanging over the AGI building is the true cause of intelligent emergence, rather than a simple interpretation based on experimental parameters.

Selected Publications

HumanNeRF- SE : A Simple yet Effective Approach to Animate HumanNeRF with Diverse Poses

Caoyuan Ma Yu-Lun Liu Zhixiang Wang Wu Liu Xinchen Liu Zheng Wang

Computer Vision and Pattern Recognition (CVPR) 2024

SH -Neus: Colored Spherical Harmonics for Neural Implicit Surface Reconstruction

Ziqiao Zhou* Caoyuan Ma* Runqi Wang Yifan Duan Zheng Wang

Under review

TAAT : Think and Act from Arbitrary Texts

Runqi Wang* Caoyuan Ma* Guopeng Li* Zheng Wang

Arxiv 2024

Improving Stack Overflow question title generation with copying enhanced CodeBERT model and bi-modal information

Fengji Zhang Xiao Yu Jacky Keung Fuyang Li Zhiwen XieZhen Yang Zhen Yang Caoyuan Ma Zhimin Zhang

Information and Software Technology 2022

All publications