Jia Xu, Ph.D.

Principal Researcher
Tencent AI Lab

Email: jiaxu [at] cs.wisc.edu
Picture of Jia Xu


I am a principal researcher at Tencent AI Lab. Before returning to China, I was a senior research scientist in the Intel Visual Computing Lab, lead by the awesome Vladlen Koltun. I received my Ph.D. in Computer Sciences at the University of Wisconsin-Madison in July 2015, with my thesis committee of Prof. Vikas Singh (advisor), Prof. Chuck Dyer, Prof. Jerry Zhu, Prof. Jude Shavlik, and Prof. Mark Craven. I was a visiting student in University of Toronto during Summer 2014, and in Toyota Technological Institute at Chicago during Summer 2013, both working with Prof. Raquel Urtasun. In the first year (2010-2011) of my graduate study, I worked as a research assistant/intern at Epic. Before graduate school, I obtained my B.S. degree from the Department of Computer Science and Technology at Nanjing University, China in June 2010.

Curriculum Vitae (pdf).

Research Interests

My major interests are computer vision, deep learning, and reinforcement learning. In particular, I am interested in systematically building reliable visual perception algorithms (e.g., semantic segmentation, dense correspondence, video analytics/segmentation) by mathematically modeling the physical properties of visual data (e.g., geometry, structure, context), and learning from the massive unlabeled or weakly labeled data available on the Internet. My ultimate research goal is to enable computers to perceive and reason at/beyond human level.

What's New

  • Dec. 2017: I joined Tencent AI Lab as a Principal Researcher. We are hiring self-motivated interns and full-time engineers/researchers. Contact me for more details.
  • July 2017: Our Fast Image Processing paper is accepted to ICCV 2017.
  • Feb. 2017: Our DCFlow paper is accepted to CVPR 2017. Winner on the Sintel Optical Flow Benchmark since 10/2016.
  • Research Projects

    Fast Image Processing

  • Fast Image Processing with Fully-Convolutional Networks
  • Optical Flow

  • Accurate Optical Flow via Direct Cost Volume Processing
  • Semantic Segmentation

  • Learning to Segment Under Various Forms of Weak Supervision
  • Tell Me What You See and I will Show You Where It Is
  • Gaze-enabled Egocentric Video Summarization

  • Gaze-enabled Egocentric Video Summarization via Constrained Submodular Maximization
  • Structured Sparsity for Video Segmentation and Spectral Clustering

  • Spectral Clustering with a Convex Regularizer on Millions of Images
  • GOSUS: Grassmannian Online Subspace Updates with Structured-sparsity
  • Interactive Segmentation and Contour Completion

  • Incorporating Topological Constraints within Interactive Segmentation and Contour Completion via Discrete Calculus
  • Cosegmentation

  • Analyzing the Subspace Structure of Related Images: Concurrent Segmentation of Image Sets
  • Random Walks based Multi-Image Cosegmentation: : Quasiconvexity Results and GPU-based Solutions

    Computer Sciences | UW Home
    Locations of visitors to this page
    eXTReMe Tracker