Xia Li

ethanlee AT pku.edu.cn

Ph.D. student

ETH Zurich

Personal Site


I am a third-year master student at Peking University (PKU). Here I am lucky enough to work with Prof. Zhouchen Lin and Prof. Hong Liu. Before studying in PKU, I obtained my bachelor’s degree from Beijing University of Posts and Telecommunications (BUPT) in 2017.


  • Computer Vision
  • Semantic Segmentation
  • Low-level Vision


  • M.S. in Computer Engineering, 2017-2020

    Peking University

  • B.S. in Network Engineering, 2013-2017

    Beijing University of Posts and Telecommunications



SWE for ML

Google GBoard Team

Apr 2019 – Jul 2019 Beijing
Worked on incremental transliteration model.

ML Intern

Bytedance AI Lab

Nov 2016 – Mar 2017 Beijing
Worked on human alpha matting.

Publications @ZERO Lab

Is Attention Better Than Matrix Decomposition?. ICLR, 2021.

As an essential ingredient of modern deep learning, attention mechanism, especially self-attention, plays a vital role in the global …

Improving Semantic Segmentation via Decoupled Body and Edge Supervision. ECCV, 2020.

Existing semantic segmentation approaches either aim to improve the object’s inner consistency by modeling the global context, or …

Spatial Pyramid Based Graph Reasoning for Semantic Segmentation. CVPR, 2020.

The convolution operation suffers from a limited receptive filed, while global modeling is fundamental to dense prediction tasks, such …

Dynamical System Inspired Adaptive Time Stepping Controller for Residual Network Families. AAAI, 2020.

Inspired from the dynamical systems, this study aims to unravel and improve ResNets.

SOGNet: Scene Overlap Graph Network for Panoptic Segmentation. AAAI, 2020.

Our study aims to explicitly predict overlap relations and resolve overlaps in a differentiable way for the panoptic output.

Expectation Maximization Attention Networks for Semantic Segmentation. ICCV, 2019.

We formulate the attention mechanism into an expectation-maximization manner and iteratively estimate a much more compact set of bases …

R^2 Net Recurrent and Recursive Network for Sparse View CT Artifacts Removal. MICCAI, 2019.

We propose a novel neural network architecture to reduce streak artifacts generated in sparse-view 2D Cone Beam Computed To-mography …

Recurrent Squeeze-and-Excitation Net for Single Image Deraining. ECCV, 2018.

We propose a novel deep network architecture based on deep convolutional and recurrent neural networksfor single image deraining.


ICCV Oral. 2019.

Oral presentation for EMANet

COCO2019. 2019.

Panoptic Segmentation Innovative Award

Recent researches on semantic segmentation. 2019.

Shares the recent researches about semantic segmentation

"I love Fashion". 2019.

The presentation of our team for Google AI ML Camp

COCO2018. 2018.

Panoptic Segmentation Runner-Up


Innovative Award (PKU_ZERO Team)

We competed on the Panoptic Segmentation track

National Scholarship

Best Projects

Runner-up (PKU_360 Team)

We competed on the Panoptic Segmentation track

Chinese Reconstruct scholarship

Beijing Marathon in 4.5 hours

Taishan Marathon in 4 hours

National Scholarship

Beijing Half-marathon in 2 hours