website page counter Xiuye's Home

XIUYE GU

Hi, I'm an AI Resident at Google Research, where I'm lucky to be advised by Yin Cui and Tsung-Yi Lin. Previously, I obtained my M.S. in Computer Science at Stanford University, and my B.E. in Computer Science at Zhejiang University. I spent half a year (9/2018-3/2019) and summer 2016 working happily with Prof. Yong Jae Lee at UC Davis. In summer 2018, I interned at TuSimple and worked on 3D point cloud scene flow estimation with Dr. Panqu Wang and Yijie Wang. At Zhejiang University, I was advised by Prof. Deng Cai working on content based image retrieval and computer vision.

My primary research interest is Computer Vision.

xiuyegu at stanford dot edu

https://github.com/laoreja


Publications

Xiuye Gu, Weixin Luo, Michael S. Ryoo, Yong Jae Lee. Password-conditioned Anonymization and Deanonymization with Face Identity Transformers. European Conference on Computer Vision (ECCV), 2020.


[paper]   [code]   [demo video]
[1-min presentation]   [10-min presentation]

Xiuye Gu, Yijie Wang, Chongruo Wu, Panqu Wang, Yong Jae Lee. HPLFlowNet: Hierarchical Permutohedral Lattice FlowNet for Scene Flow Estimation on Large-scale Point Clouds. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019.


[paper]   [supp]   [code]   [poster]   [video]

Maheen Rashid, Xiuye Gu, Yong Jae Lee. Interspecies Knowledge Transfer for Facial Keypoint Detection. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.


[paper]   [code]   [anotation tool]

Deng Cai, Xiuye Gu, Chaoqi Wang. A Revisit on Deep Hashing for Large-scale Content Based Image Retrieval. Technical report, arXiv:1711.06016, 2017.


[arxiv]

Projects

Explore Deep Graph Generation


Course project for CS224W: Machine Learning with Graphs.
Explore deep graph generation from two directions:
1) use CNN GANs to model the whole adjacency matrix directly after sorting the nodes;
2) build upon the very recent Graph Recurrent Attention Networks (GRANs), proposed a graph completeness judger network and improved its attention mechanism.

[poster]   [report]   [code]

Simulating and Rendering Explosion


Course project for CS348B: Image Synthesis Techniques.
Extended PBRT to support emissive volumes and openVDB input format.
Self-studied and implemented blackbody radiation, closed-form and delta tracking.
Simulated explosion and flying rubbles using Blender.

[report]   [video]   [code]

License Plate Detection and Character Segmentation


A robust iterative license plate character segmentation algorithm and a license detection system with robust skew and slant correction to improve character segmentation.

[character segmentation report]   [detection report]
[detection survey]   [segmentation survey]   [recognition survey]
[code]   (reports and surveys are in Chinese)

Deep Stereo Matching


Course project for CS231A: Computer Vision, From 3D Reconstruction to Recognition.

[code]   [report]   [supp]