Zuodong Zhong (钟祚栋)
Ph.D. StudentPeking University
I am a Ph.D. student at Peking University, advised by Asst. Prof. Yixing Huang. I earned my B.Eng. (2026) in Automation from the University of Science and Technology Beijing, where I was advised by Prof. Qing Li.
My research focuses on efficient architectures for long-video understanding, particularly sparse attention mechanisms. I aim to help multimodal large language models process long videos more efficiently and effectively. I am fortunate to receive research guidance from Assoc. Prof. Longteng Guo and Prof. Jing Liu.
I previously interned with the Super Intelligence Team at Xiaohongshu, where I was a core contributor to FireRed-OCR.
During an earlier internship at the Institute of Automation, Chinese Academy of Sciences (CASIA), I studied diffusion models and later vision-language models (VLMs) for autonomous driving, and contributed to ReasonPlan. I was advised by Assoc. Prof. Qichao Zhang and Prof. Dongbin Zhao.
I enjoy using the Feynman technique and share my notes on diffusion models and research on my blog and Zhihu column.
Simple and EffectiveMy research philosophy.
Contact me on WeChat
Scan with WeChat, or click the image to view it full size.
News
-
FireRed-OCR is now open source! At release, it achieved SOTA among end-to-end methods on OmniDocBench v1.5.
-
ReasonPlan has been accepted to the Conference on Robot Learning (CoRL 2025)!
Publications & Projects
-
FireRed-OCR- A 2B-parameter vision-language model for end-to-end structured document parsing, combining progressive training with format-constrained reinforcement learning. Achieved 92.94 on OmniDocBench v1.5, reaching SOTA among end-to-end methods at release.
-
ReasonPlan: Unified Scene Prediction and Decision Reasoning for Closed-loop Autonomous Driving- Combining next-scene prediction and decision chain-of-thought to improve multimodal reasoning for closed-loop autonomous driving.
Blogs
All posts →-
DDPM
Adding noise in one step, denoising step by step
-
DDIM
The same marginals, fewer sampling steps
-
Score-based Generative Models
Understanding diffusion and denoising through gradients
Research Experience
-
Super Intelligence Team, Xiaohongshu
November 2025 – April 2026, Research Intern.
Core contributor to FireRed-OCR, researching end-to-end structured document parsing.
-
Institute of Automation, Chinese Academy of Sciences
July 2024 – August 2025, Research Intern.
Explored diffusion models and vision-language models for autonomous driving; contributed to ReasonPlan.
Advised by Assoc. Prof. Qichao Zhang and Prof. Dongbin Zhao.
Education
-
-
University of Science and Technology Beijing
B.Eng. in Automation, School of Automation and Electrical Engineering, August 2022 – June 2026
Served as President of the Suo'ao Technology Center student club.
Advised by Prof. Qing Li.
Selected Honors & Awards
- Outstanding Graduate of Beijing Higher Education Institutions
- Baosteel Outstanding Student Award (awarded to only two undergraduates at USTB each year)
- RAICOM National Finals, Second Prize
- Yonggang Scholarship (Top 3%)