Zeng Huang
Ph.D. candidate at USC 黄锃

Zen photo
Univ. of Southern California
PHE 108, 3737 WATT WAY
90089 Los Angeles

I'm a Ph.D. student in Computer Science, focusing on research of Computer Vision and Graphics, and aiming at building the bridge between sciences and arts.
I am also a skilled drawer, painter, photographer and camera expert.

 

Education

Graduate Ph.D.

USC
2016 - Present

Ph.D., Computer Science

CS Dept. Viterbi School of Engineering, advised by Dr. Hao Li
University of Southern California, U.S.
Undergraduate

SJTU
2012 - 2016

B.Eng., Computer Science and Technology

ACM Honored Class (a pilot computer science class in China), Zhiyuan College
Shanghai Jiaotong University, China
 

Publications

ECCV
2018

DEEP VOLUMETRIC VIDEO FROM VERY SPARSE MULTI-VIEW PERFORMANCE CAPTURE

Zeng Huang, Tianye Li, Weikai Chen, Yajie Zhao, Jun Xing, Chloe LeGendre, Linjie Luo, Chongyang Ma, Hao Li

ICLR
2018

AUTO-CONDITIONED LSTM NETWORK FOR EXTENDED COMPLEX HUMAN MOTION SYNTHESIS

Zimo Li, Yi Zhou, Shuangjio Xiao, Chong He, Zeng Huang, Hao Li

ICCV
2017

Realistic Dynamic Facial Textures from a Single Image using GANs

Kyle Olszewski, Zimo Li, Chao Yang, Yi Zhou, Ronald Yu, Zeng Huang, Sitao Xiang, Shunsuke Saito, Pushmeet Kohli, Hao Li

 

Honors and Awards

Fellowship

USA
2016

USC Annenberg Graduate Fellowship

'Fellows will conduct cutting-edge communication and digital media research, advance important new programs in the communications arena, and constitute an internationally recognized and highly regarded group of communication research scholars and creative practitioners.'

Work

China
2016

Star of Tomorrow

In recognition of participation in the Microsoft Research Asia Internship Program.
An interview on me.

Competition

USA
2014

Meritorious Winner

First Prize in Mathematical Contest In Modeling, 2014
Paper accepted by Science and Technology & Innovation, Shanxi, China
I introduced a way of using Cellular automaton to simulate real-time traffic flow and implement it in C++. In this simulation, we successfully produced behaviors of multiple traffic rules, and observed an interesting phenomenon of chaos.

Scholarship

China
2013, 2014, 2015

Academic Excellence Scholarship

Undergraduate scholarship in consecutive three years, Shanghai Jiaotong University

Competition

China
2012

First Prize

National High School Physics Olympiad, China
With this award, I got the chance to skip the High School Entrance Examination in China. I was directly promoted to college.

 

Teaching Experiences

TA

2014 - 2015

Teaching Assistant

Data Structures, 2014
My job was to teach extended knowledge of coding and algorithm skills. I also organized a self-learning group for those who had no coding experience before college, which helped them quickly catch up with others. Besides, I helped to carry out tests and quizzes.

Principle and Practice of Computer Algorithms , Summer 2014
My job was to teach front-end web design to a group of interested students. I also helped them to redesign the class's website.

 

Research/Work Experiences

Computer Graphics

Present

Research Intern

at Facebook Inc.'s Oculus VR
Advisor: Dr. Secret

Mission undergoing.

Computer Graphics

Present

Ph.D. Candidate

at University of Southern California
Advisor: Dr. Hao Li

Topics: Performance Capturing and Deep Learning.
I am currently working at University of Southern California as a Ph.D. student under the supervisor of Dr. Hao Li. My current research topic mainly focuses on human body capturing and high-quality reconstruction from multi-view video clips, using low-cost family-avaible equipments.
I also work with the graphics lab at the Institute for Creative Technologies.

Computer Vision

2015-2016

Research Intern

in the Internet Media Group, at Microsoft Research Asia
Advisor: Dr. Wenjun Zeng, Cuiling Lan, Junliang Xing

Topics: Human Detection and Pose Estimation.
Human Detection locates human body in a given image. Pose Estimation then determines positions of each body parts. With this information, further work, such as Pose Classification can be carried out.
For Human Detection, I worked on porting RCNN and Faster-RCNN to the specified problem on humans. I tried multipal data argument methods and practiced adjusting learning parameters, to get better results. I have been supporting our group with this detection framework that was further used in many related works.
Laterly I turned my focus to Pose Estimation. Following previous work of Deeppose and Fully-convolutinal Neural Network, I proposed a network that takes in arbitrary input image size and produces both pixel-wise appearance information for each human part and fixed-length regression of joints positions. After then, I tried extracting parts correlation from regression to refine appearance with a self-correction strategy and got a result with 75% PCK comparative to the state-of-art.
A demo of face bluring with my work for head detecting.

Deep Learning

2015

Visiting Student

at Cornell University
Advisor: Prof. John Hopcroft

Topics: Convolutional Neural Network.
Deep Learning is a new area of Machine Learning, consists of multipal levels of representation and abstraction. Having achieved much success and development, many strategies in Deep Learning still lacks novel explaination.
Cornell offered us a good opportunity to have a month of wide discussion on the current development of Deep Learning methods, such as the network's representative ability, sparsity, and the importance of the data scale. I worked on the survey of the CNN Network. Typically, I was most interested in the represetation of the hidden gates in the deep structure, and I tried implementing and visualizing the network. I also gave a talk on the basic ideas of CNN Network.

Computer Vision

2014 - 2015

Research Assistant

at Brain-like Computing and Machine Intelligence Lab, Shanghai Jiaotong University
Advisor: Prof. Hongtao Lu

Topics: Stereo Vision and Optical Flow.
Stereo Vision recovers depth information from a pair of images of the same scene, while Optical Flow tracks the movement of pixels in adjacent video frames. Both of them call for pixel-level Image Matching to produce correspondence information.
Corespondence matching can be seen as a global optimization problem. I fomulate the matching costs as gravitational force and the image appearance as spring-network. By this mean, the matching problem is turned into solution of a motion system. Using simulation, I observed a morphing between each pair of images and got a result comparative to the traditioal methods.

 

Undergraduate Projects

Web

2014 - 2015

0CTF, HTML CSS JS

International Security Competition
As director of the web server front-end design and implementation

Operating System

2014

Nachos Operating System, Java

Nachos Project from UC Berkeley CS162

Shell

2014

Unix-like shell, C

Implement a shell supporting job control

Compiler

2013 - 2014

Modern Compiler Implementation, C

Compiler for Simplified C Language
The course-project is difficult as is written in C with no Object Oriented support, few classmates tried this way (they use Java with supporting classes)
This work is full-scored with 100 points

Computer System

2013 - 2014

Simulated GPGPU, C++

GPGPUsim Project from University of British Columbia

Graphics

2013

EvoLisa, C++

Implement the interesting evolution algorithm for image triangularization providing artistic works

 

Activities

Photography

2013.06 - 2013.12

President

of Student Photography Association, Shanghai Jiaotong University
Leader of photography in university, invited to many top activities as journalist

Literature

2013.08

Co-founder

of Southwester, a student magazine, Shanghai Jiao Tong University
当你有梦 而我们敢想【西南风·创刊】

 

Reference

 

You may refer to my personal website, github page and portfolio for details.