Zechuan Zhang (张泽川)

I am currently a PhD student at Zhejiang University in Hangzhou, China. My PhD supervisor is Prof. Yi Yang. My research interests lie in the intersection of computer vision, machine learning. I am particularly interested in 3D vision, multi-modal, diffusion models and image generation and editing.

Prior to that, I obtained the B.Sc Degree in Geographical Information Science from Zhejiang University in 2023. I was also a member of Advanced Honor Class of Engineering Education (ACEE) at Chu Kochen Honors College (CKC) of Zhejiang University.

Email  |  CV  |  GitHub  |  Google Scholar  |  Twitter

headshot
News
  • 06/2025 I've joined the Broad Institute of Harvard and MIT as a visiting scholar, working with Prof. Siwei Chen and Prof. Benjamin Neale.
  • 04/2025 We released ICEdit, a novel framework for instruction-guided image editing.
  • 10/2024Achieved First Prize (Top 0.2%) in the China Graduate Artificial Intelligence Competition, ranking 4th among 2,228 teams across the country.
  • 04/2024 Our paper SIFU is selected as Highlight (top 12%) by CVPR2024!
  • 04/2024 Invited to give a talk (in Chinese) on Real-world Usable Clothed Human Reconstruction at SHUZIHUANYU.
  • 03/2024 One paper was accepted by CVPR 2024!
  • 09/2023 One paper was accepted by NeurIPS 2023!
    • ---- show more ----
Research
Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer
Zechuan Zhang, Ji Xie, Yu Lu, Zongxin Yang, Yi Yang
ArXiv, 2025
paper | project page | code | GitHub stars

We introduce ICEdit, an efficient and effective framework for instruction-based image editing. With only 1% trainable parameters (200M) and 0.1% training data (50k) compared to previous methods, ICEdit demonstrates strong generalization capabilities and is capable of handling diverse editing tasks. Compared with commercial models such as Gemini, GPT4o, we are more open-source, with lower costs, faster speed (it takes about 9 seconds to process one image), and powerful performance.

3D object manipulation in a single image using generative models
Ruisi Zhao, Zechuan Zhang, Zongxin Yang, Yi Yang
ArXiv, 2025
paper | project page

We introduce OMG3D, a novel framework that combines precise geometric control with the generative power of diffusion models, thus achieving significant improvements in visual performance.

SIFU: Side-view Conditioned Implicit Function for Real-world Usable Clothed Human Reconstruction
Zechuan Zhang, Zongxin Yang, Yi Yang,
Conference on Computer Vision and Pattern Recognition (CVPR), 2024 (Highlight, top 12%)
paper | project page | poster | code | media (AK) | [新智元]| GitHub stars

With just a single image, SIFU is capable of reconstructing a high-quality 3D clothed human model, making it well-suited for practical applications such as scene creation and 3D printing.

Global-correlated 3D-decoupling Transformer for Clothed Avatar Reconstruction
Zechuan Zhang, Li Sun, Zongxin Yang, Ling Chen, Yi Yang
Conference on Neural Information Processing Systems (NeurIPS), 2023
paper | project page | code | media (CVer)| GitHub stars

A transformer-based implicit function for single image clothed human reconstruction.

Academic Services
  • Conference Reviewer: ACM MM, NeurIPS, Siggraph Asia
  • Journal Reviewer: TPAMI
Awards
  • First Prize in the China Graduate Artificial Intelligence Competition (4/2228).
  • BYD Scholarship in the session of 2024.
  • CHU KOCHEN Scholarship in the session of 2022 (Highest Honor at Zhejiang University).
  • China National Scholarship for undergraduate in the session of 2020, 2021 and 2022.
  • First Class Scholarship of Zhejiang University in the session of 2020, 2021 and 2022.
  • Outstanding Winner in National Application of GIS Skill Competition in 2021.

template adapted from this awesome website