About me

Hi, I recently graduated from Carnegie Mellon University (CMU) with M.S. in Computer Science and became a new research associate in the Robotics Institute, where I am advised by Dr. Kris M. Kitani. Before coming to CMU, I received my B.S. in Electrical and Computer Engineering at Hanyang University, Seoul, Korea, where I had a chance of working with Prof. Jongwoo Lim. I was also a visiting student at the University of California, Los Angeles (UCLA) and Drexel Univeristy in 2008 and 2012 respectively. My graduate study is fully supported by Graduate Study Abroad Scholarship from Jeongsong Cultural Foundation, Korea.

Research interests

I am interested in solving problems in Computer Vision using Machine Learning techniques as well as developing real-world applications. In particular, my current research focus is on developing structured deep networks well-fitted for multi-task problems.

*Area of interests: Computer Vision, Computational Photography, Machine Learning, and Robotics

*Ongoing projects: 1) Human pose estimation by synthesis, 2) Human activity forecasting



Forecasting Pedestrians in Urban Environment
: Forecast interactive dynamics of multiple people

Key techniques: Convolutional neural network, Inverse reinforcement learning
*Paper: A Game-Theoretic Approach to Multi-Pedestrian Activity Forecasting, ECCV16 (In submission)

Sports Activity Forecasting
: Predict viable paths for an agent given a single image

Key techniques: Inverse reinforcement learning, Gaussian proceess regression
*Paper: Predicting Wide Receiver Trajectories in American Football, WACV16 (Oral)

Food Sensor
: Identify food type, calculate volume and measure calories

Key techniques: Convolutional neural network, Point cloud registration *Developed in part as a project for the UAE Robotics Competition, 2016

Dense 3D Human Reconstruction in the Wild
: Reconstruct a dense 3D human

Key techniques: Non-rigid dense correspondences, Human pose matching via a large 3D motion database (CMU panoptic studio)
*Best presentation award (Learning Based Method in Vision, CMU, Spring 2015)

Exploiting Feature Statistics with Multi-correspondences
: Find more inlier matches fast

Key techniques: Feature orientation, Multi-correspondences
*Paper: Improving Consensus Set Maximizaion via Feature Statistics with Multi-correspondences (In preparation)