Xuhai Chen

I'm a Master's student at Zhejiang University in China, advised by Prof. Yong Liu, and expect to graduate in 2026.

In 2024, I interned at the State Key Laboratory of CAD&CG at Zhejiang University, where I worked on motion generation with Prof. Xiaowei Zhou and Prof. Sida Peng. In 2025, I worked as a remote research intern at the University of Pennsylvania, focusing on 3D reconstruction and generation with Prof. Lingjie Liu.

My current research focuses on 3D vision, particularly motion generation and 3D reconstruction and generation. I have also worked on image super-resolution and anomaly detection.

I enjoy exploring new experiences and challenges, which makes me feel happy and fulfilled. Life is meant to be experienced.

Email / Scholar / Github / CV

Awards

[2023.06]: CVPR 2023 workshop VAND Challenge: Winner in the Zero-shot Track, Honorable Mention in the Few-shot Track.

Papers

	Contact Matrix: Enhancing Dance Motion Synthesis with Precise Interaction Modeling Xuhai Chen, Zhi Cen, Huaijin Pi, Sida Peng, Xiaowei Zhou, Yong Liu CVPRF, 2026 paper / bibtex Generate realistic reactive motions for duet dance by modeling fine-grained leader–follower contacts with contact-guided diffusion.
	Better "CMOS" Produces Clearer Images: Learning Space-Variant Blur Estimation for Blind Image Super-Resolution Xuhai Chen, Jiangning Zhang, Chao Xu, Yabiao Wang, Chengjie Wang, Yong Liu CVPR, 2023 paper / github / bibtex Estimating space-variant blur degradation with the help of semantic information.
	A Zero-/Few-Shot Anomaly Classification and Segmentation Method for CVPR 2023 VAND Workshop Challenge Tracks 1&2: 1st Place on Zero-shot AD and 4th Place on Few-shot AD Xuhai Chen, Yue Han, Jiangning Zhang arXiv, 2023 paper / github / bibtex Technical report for the VAND challenge at the 2023 CVPR workshop.
	CLIP-AD: A Language-Guided Staged Dual-Path Model for Zero-shot Anomaly Detection Xuhai Chen, Jiangning Zhang, Guanzhong Tian, Haoyang He, Wuhao Zhang, Yabiao Wang, Chengjie Wang, Yong Liu arXiv, 2023 paper / github / bibtex Adapt the CLIP model for anomaly segmentation by merely fine-tuning a linear layer, and explain the text prompts design from a distributional perspective.
	Exploring Plain ViT Reconstruction for Multi-class Unsupervised Anomaly Detection Jiangning Zhang, Xuhai Chen, Yabiao Wang, Chengjie Wang, Yong Liu, Xiangtai Li, Ming-Hsuan Yang, Dacheng Tao arXiv, 2023 paper / github / bibtex Construct a reverse distillation architecture for multi-class anomaly detection using plain ViT.