Xueyan Zou

xueyan@cs.wisc.edu

About

I am currently a PhD student at Computer Science Department in University of Wisconsin, Madison, supervised by Prof. Yong Jae Lee. Prior to this, I spent three wonderful years at University of California, Davis, advised by the same supervisor and worked closely with Dr. Fanyi Xiao. Previously, I received bachelor's Degree at Hong Kong Baptist University where I was advised by Prof. PC Yuen and Prof. Xiaowen Chu.

Research Interest: Object/Instance/Edge Detection, Video Analysis, Instance Relationship Reasoning, AI&Art

Experience

Publications

Progressive Temporal Feature Alignment Network for Video Inpainting

Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021
Xueyan Zou , Linjie Yang, Ding Liu, Yong Jae Lee

Delving Deeper into Anti-Aliasing in ConvNets

Proceedings of the British Machine Vision Conference (BMVC), 2020
Xueyan Zou , Fanyi Xiao, Zhiding Yu, Yong Jae Lee
Best Paper Award, Oral Presentation

Previous Projects

Unsupervised style transfer and object transfiguration

In this work, we explore unsupervised Image-to-Image translation, proposing a single network could handle style transfer and object transfiguration. In addition, We introduce a learning phase called "identity pre-train”, which helps to stabilize training on difficult object transfiguration task.

A probability view of different generative adversarial net

In this work, I visualize the probability distribution shifting proceduring when training GANs using data points from sampled gaussian distributions. To have a better understanding of different GANs dynamic, I compare the dynamics among GAN, simGAN and cycleGAN.

An improved Cooperative and Penalized Competitive Learning Algorithm

CPCL, a clustering alorithm could find the cluster centroid and number without knowing the data distribution. However its intra-cluster converging speed is slower than the inter-cluster converging speed. The proposed algorithm accelerate and increase its accuracy using a signed network approach.

Traffic Prediction Using LSTM Network

In this project, the task is to predict the traffic flow in Beijing on 2016/04/20 from 8:00AM-10:00AM. Different Network structures and data preprocessing procedure were tried in this project. During the competition, performance was improved from 0.8086 (Baseline) to 0.5423 on RMSE, ranked 4th over 230 teams worldwide.

Misc

  • Reviewer: CVPR22, BioInfomatics. (Draft reviews on NeurIPS2019/2020, BMVC2021)