Zichen Zou

Zichen Zou (邹子辰)

Ph.D. student at Fudan University

Welcome to Zichen Zou’s (邹子辰) personal website!

I am currently a first-year Ph.D. student (direct Ph.D. track) at the College of Computing and Intelligent Innovation, Fudan University, advised by Prof. Xiaosong Jia (贾萧松). Previously, I received my B.Eng. in Artificial Intelligence (Excellent Talent Pilot Class) from Shanghai Jiao Tong University (SJTU), where I was a member of the Zhiyuan Honors Program (an elite program for top 5% students) and worked closely with Prof. Yulun Zhang.

My research focuses on feed-forward 3D reconstruction and novel view synthesis. Recently, I am also broadly interested in world models, especially 3D world generation and reconstruction from multi-modal inputs.

I am always open to collaborations and academic discussions. Feel free to reach out via email at zzcnb123456@sjtu.edu.cn. You can also find me on Xiaohongshu (小红书): 周小邹 (ID: 49891398784).

🔥 News

  • 2026.09:  🎓 I joined FVL Lab (Fudan Vision and Learning), Fudan University, and started my Ph.D. journey, advised by Prof. Xiaosong Jia!
  • 2026.09:  🎉🎉 Our paper RetrieveVGGT has been accepted to NeurIPS 2026!
  • 2026.06:  🏅 I was awarded Outstanding Graduate of Shanghai Jiao Tong University (上海交通大学优秀毕业生)!
  • 2026.02:  🎉🎉 Our paper Efficient-LVSM has been accepted to ICLR 2026!
  • 2025.09:  🎉🎉 Our papers DOVE and Distillation-Free One-Step Diffusion have been accepted to NeurIPS 2025!
  • 2024.09:  🔬 I joined Prof. Yulun Zhang’s group at Shanghai Jiao Tong University, working on efficient diffusion models for image/video restoration.

📝 Publications

NeurIPS 2026
RetrieveVGGT framework

Attention Itself Could Retrieve. RetrieveVGGT: Training-Free Long Context Streaming 3D Reconstruction via Query-Key Similarity Retrieval

Zichen Zou, Xiaosong Jia, Zuxuan Wu, Yu-Gang Jiang

Project · Code

  • TL;DR: A training-free streaming 3D reconstruction framework that retrieves relevant history via query-key similarity, enabling long-context reconstruction without fine-tuning.
NeurIPS 2025
DOVE pipeline

DOVE: Efficient One-Step Diffusion Model for Real-World Video Super-Resolution

Zheng Chen†, Zichen Zou†, Kewei Zhang, Xiongfei Su, Xin Yuan, Yong Guo, Yulun Zhang*

Project · Code

  • TL;DR: An efficient one-step diffusion model for real-world video super-resolution. Project page has received 30,000+ visits and the code repository 200+ GitHub stars.
ICLR 2026
Efficient-LVSM structure

Efficient-LVSM: Faster, Cheaper, and Better Large View Synthesis Model via Decoupled Co-Refinement Attention

Xiaosong Jia, Yihang Sun, Junqi You, Songbur Wong, Zichen Zou, Junchi Yan, Zuxuan Wu, Yu-Gang Jiang

Project · Code

  • TL;DR: Decoupled co-refinement attention makes large view synthesis models faster and cheaper while improving reconstruction quality.
NeurIPS 2025
DFOSD overview

Distillation-Free One-Step Diffusion for Real-World Image Super-Resolution

Jianze Li, Jiezhang Cao, Zichen Zou, Xiongfei Su, Xin Yuan, Yulun Zhang, Yong Guo, Xiaokang Yang

Code

  • TL;DR: A distillation-free training strategy that turns diffusion models into one-step generators for real-world image super-resolution.

🎖 Honors and Awards

  • 2026.06: Outstanding Graduate of Shanghai Jiao Tong University (上海交通大学优秀毕业生)
  • 2022–2025: Zhiyuan Honors Scholarship (致远荣誉奖学金, Top 5%), SJTU, four consecutive years

📖 Educations

  • 2026.09 - now, Ph.D. student (direct Ph.D. track), College of Computing and Intelligent Innovation, Fudan University, Shanghai, China. Advisor: Prof. Xiaosong Jia.
  • 2022.09 - 2026.06, B.Eng. in Artificial Intelligence (Excellent Talent Pilot Class), Shanghai Jiao Tong University, Shanghai, China. Zhiyuan Honors Program (top 5%). GPA 3.90/4.30, ranked 23/94.