Siyuan Huang (黄思远)

I am a senior year undergraduate student in Department of Automation, School of Information Science and Technology, Tsinghua University.

I worked with Prof. Jiwen Lu and Prof. Jie Zhou in Institute of Information Processing, Tsinghua University since 2014. I visited Prof. Anil K. Jain at PRIP Lab, Michigan State University in the summer 2015.

My research interests lie in computer vision and machine learning (metric learning, deep learning, etc).

I will graduate from Tsinghua University in July 2016 and I am applying for PhD starting from Fall 2016. If you have interest in me, please don't hesitate to contact me.


Publications

Nonlinear local metric learning for Person Re-identification
Siyuan Huang, Jiwen Lu, Jie Zhou, Anil K. Jain
Submitted to IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, June 26 - July1, 2016
[paper]

Image-Set Querying Based Localization (oral)
Lei Deng*, Siyuan Huang*, Jiwen Lu, Jie Zhou, Baohua Chen (* indicates equal contributions)
IEEE Visual Communication and Image Processing (VCIP), Singapore, December 13-16, 2015
[paper]

Building Change Detection Based on 3D Reconstruction
Baohua Chen, Lei Deng, Yueqi Duan, Siyuan Huang, Jie Zhou
IEEE International Conference on Image Processing (ICIP), Quebec City, Canada, September 27-30, 2015
[paper]
Back home

Education
2012 - 2016 (Expected), Department of Automation, Tsinghua University,

Balchlor of Engineering

Major GPA: 91.4/100

Back home

Research  

Nonlinear local metric learning for Person Re-identification
Siyuan Huang, Jiwen Lu, Jie Zhou, Anil K. Jain
Submitted to IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016

  • We propose a nonlinear local metric learning (NLML) method to improve the state-of-the-art performance of person re-identification on public datasets. Motivated by the fact that local metric learning has been introduced to handle the data which varies locally and deep neural network has presented outstanding capability in exploiting the nonlinearity of samples, we utilize the merits of both local metric learning and deep neural network to learn multiple sets of nonlinear transformations.

  • By enforcing a margin between the distances of positive pedestrian image pairs and distances of negative pairs in the transformed feature subspace, discriminative information can be effectively exploited in the developed neural networks.

  • Our experiments show that the proposed NLML method achieves the state-of-the-art results on the widely used VIPeR, GRID, and CUHK 01 datasets.

[paper]

Image-Set Querying Based Localization (oral)
Lei Deng*, Siyuan Huang*, Jiwen Lu, Jie Zhou, Baohua Chen (* indicates equal contributions)
IEEE Visual Communication and Image Processing (VCIP), 2015

  • Developed a new optimization method to estimate the locations and poses of cameras based on image sets.

  • Used bridging images to help the target image reconstruct a local 3D model and match the 3D point cloud of the area of interest, and performed nonlinear optimization.

  • Built scean 3D point cloud for the Main Building in Tsinghua University for experiments and results show the proposed method solves a lot of cases where conventional single image based localization method cannot estimate the poses of cameras.

[paper]

Building Change Detection Based on 3D Reconstruction
Baohua Chen, Lei Deng, Yueqi Duan, Siyuan Huang, Jie Zhou
IEEE International Conference on Image Processing (ICIP), 2015

  • Implemented building change detection by reconstructing the local 3D model of target area online and compared it with the 3D model reconstructed offline.

  • Got the depth map of target area by utilizing Structure-from-Motion algorithm to reconstruct the 3D scenes.

  • The ROC curve indicates superior quality of the generated difference depth map for building change detection compared to other traditional correlation analysis methods based on radiometric images.

[paper]

Car Detection Based on 3D Reconstruction
Siyuan Huang, Jie Zhou

  • Solved the problem of dual-view and wide baseline car detection by combining object detection and Structure-from-Motion.

  • Used multiple uncalibrated cameras to reconstruct the 3D image scenes and matched the cars detected from different cameras to solve car blocking problem.

  • Improved object detection results given multiple uncalibrated images. Fewer input and easy way to get 3D transformation.

[pdf]

Prediction of the Reliability of Face Verification
Siyuan Huang, Cheng Li
Computer vision group of SenseTime

  • Designed damaging operators such as Gaussian Blur and Motion Blur to simulate the face images with different levels of quality.

  • Developed the quality classifier to learn the relationship between face verification result and image quality, which is used to evaluate the reliability of face verification.

Facial Expression Intensity Detection in Video Streams
Siyuan Huang, Cheng Li
Computer vision group of SenseTime

  • Explored facial expression intensity by jointly learning the face and text information in video clips.

  • Built a new dataset of facial expressions with more than 10,000 images by detecting, tracking andextracting the intense expressions in video streams automatically.

Real-time Multiple Face Tracking
Siyuan Huang, Cheng Li
Computer vision group of SenseTime

  • Applied RANSAC algorithm to accelerate feature-matching module and achieved real-time multiple face tracking on mobile devices at high speed accurately.

  • Optimized the Detection and Tracking modules in Face Software Development Kit and applied it to achieve competitive performance in visual surveillance.

[demo1 in product] [demo2 in product]

Back home

Awards and Honors  
  • Scholarship of Excellent Academic Performance, Tsinghua University, 2015
  • Scholarship of Excellent Academic Performance, Tsinghua University, 2014
  • HAGE Scholarship, Department of Automation, Tsinghua University, 2014
  • Comprehensive Merit Scholarship, Tsinghua University, 2013
  • Second Prize in the National Physics Contest of College Students, 2013
  • Advanced Person and Team Leader of Social Practice, Tsinghua University, 2013
Skills and Interests  
  • Experienced in C, C++ and MATLAB
  • Famliliar with Python, C#, Opencv, Linux, LATEX
  • Good at designing algorithm
  • Familiar with computer vision and machine learning algorithms
  • Adept at Mathmetics and Physics
  • Basketball fan and music lover
Back home

Contact  

Email: huang-sy12@mails.tsinghua.edu.cn

Phone: (+86) 18811583731

Address: Room 520B, Zijing Building 2, Tsinghua University, Beijing, P.R. China

Back home

Copyright © Siyuan Huang 2015. All rights reserved.