Haiyang Xu ｜ 徐海阳

Haiyang is a first-year Ph.D student at University of California, San Diego (UCSD) since Fall 2024, supervised by Prof. Zhuowen Tu. Previously, he received his B.S. degree in Data Science from University of Science and Technology of China (USTC) in 2024.

His research focus is on generative models and multimodal learning, especially on topics such as controllable generation and vision language models. He is currently working with Prof. Saining Xie at NYU Courant on agent-based generative models.

Feel free to drop him an email to dicuss anything about research, life, and ourselves!

Email  /  Scholar  /  X (Twitter)  /  Github

profile photo

News

[2024.06] 🎉 Our work FeatureMixture is finally accepted by TIP 2024! 🎉
[2024.04] 🎉 Our work OmniControlNet is accepted by CVPR Workshop 2024! 🎉
[2024.02] 🎉 Our work BDM is accepted by CVPR 2024! 🎉

Publications
Bayesian Diffusion Models for 3D Shape Reconstruction
Haiyang Xu*, Yu Lei*, Zeyuan Chen, Xiang Zhang, Yue Zhao, Yilin Wang, Zhuowen Tu
CVPR, 2024
project page / paper / code

We propose Bayesian Diffusion Model (BDM) for 3D shape reconstruction, which performs effective Bayesian inference by tightly coupling the top-down (prior) information with the bottom-up procedure via joint diffusion processes.

OmniControlNet: Dual-stage Integration for Conditional Image Generation
Yilin Wang*, Haiyang Xu*, Xiang Zhang, Zeyuan Chen, Zhizhou Sha, Zirui Wang, Zhuowen Tu
CVPR Workshop, 2024
paper

We propose OmniControlNet, a dual-stage integration framework for conditional image generation, which could effectively generate different controls and corresponding images in a unified framework.

Feature Mixture on Pre-Trained Model for Few-Shot Learning
Shuo Wang, Jinda Lu, Haiyang Xu, Yanbin Hao, Xiangnan He
TIP, 2024

We propose a new feature mixture mechanism to improve the context extraction ability of the pre-trained model for few-shot learning.

Great template from Jon Barron.