Sungheon Park

I am a research scientist at Samsung Advanced Institute of Technology (SAIT) AR & Graphics Lab where I am working on the intersection of 3D vision and graphics. Prior to joining SAIT, I completed my PhD at Machine Intelligence and Pattern Analysis Lab, Seoul National University, Korea.

Email  /  CV  /  Google Scholar  /  Github

profile photo
Research

My research interest lies in computer vision and deep learning, with emphasis on 3D pose estimation and reconstruction of dynamic objects and scenes.

Procrustean Regression Networks: Learning 3D Structure of Non-Rigid Objects from 2D Annotations
Sungheon Park*, Minsik Lee*, Nojun Kwak
ECCV, 2020
arXiv / video / code

Non-rigid structure from motion cost function is directly applied as a loss function of neural nets to learn 3D shapes from 2D inputs.

Pose estimator and tracker using temporal flow maps for limbs
Jihye Hwang, Jieun Lee, Sungheon Park, Nojun Kwak
IJCNN, 2019 (Honorable Mention Award in the POSETRACK challenge at ECCV 2018 workshops)
arXiv

A pose estimator and tracker based on Limb Flow Maps.

3D Human Pose Estimation with Relational Networks
Sungheon Park, Nojun Kwak
BMVC, 2018
arXiv / video / code

3D human pose estimator from 2D joint positions with robustness to missing joints.

Music Source Separation Using Stacked Hourglass Networks
Sungheon Park, Taehoon Kim, Kyogu Lee, Nojun Kwak
ISMIR, 2018
arXiv / video / code

CNNs for human pose estimation even works well with music source separation.

Procrustean Regression: A Flexible Alignment-Based Framework for Nonrigid Structure Estimation
Sungheon Park, Minsik Lee, Nojun Kwak
Transactions on Image Processing, 2018
paper

A non-rigid structure from motion algorithm that incorporates rotation alignment, which can be optimized via gradient-descent based methods.

Independent Component Analysis by lp-norm Optimization
Sungheon Park, Nojun Kwak
Pattern Recognition, 2018
paper

Independent component analysis can be implemented by minimizing or maximizing Lp norms.

Athlete pose estimation by a global-local network
Jihye Hwang, Sungheon Park, Nojun Kwak
CVPR CVSports workshop, 2017
paper

A 2D human pose estimator that combines global and local information.

Analysis on the Dropout Effect in Convolutional Neural Networks
Sungheon Park, Nojun Kwak
ACCV, 2016
paper / code

New regularization method for convolutional layers named Max-drop was proposed.

3D Human Pose Estimation Using Convolutional Neural Networks with 2D Pose Information
Sungheon Park, Jihye Hwang, Nojun Kwak
ECCV Geometry Meets Deep Learning Workshop, 2016
arXiv

2D pose estimation results are directly used for 3D pose estimation.

Cultural Event Recognition by Subregion Classification with Convolutional Neural Network
Sungheon Park, Nojun Kwak
CVPR ChaLearn Looking at the People Workshop, 2015 (3rd place in cultural event classification challenge)
paper

Image classification by classifying region proposals.

Illumination Robust Optical Flow Estimation by Illumination-Chromaticity Decoupling
Sungheon Park, Nojun Kwak
ICIP, 2015
paper

TV-L1 optical flow optimization on HSL color space.

Polyp Detection in Colonoscopy Videos Using Deeply-Learned Hierarchical Features
Sungheon Park, Myunggi Lee, Nojun Kwak
Technical Report, 2015
paper

Participated in polyp detection challenge at ISBI 2015 workshops.

Line-based Single View 3D Reconstruction in Manhattan World for Augmented Reality
Sungheon Park, Hyeopwoo Lee, Suwon Lee, Hyun S. Yang
VRCAI, 2015
paper
Smartphone as an Augmented Reality Authoring Tool via Multi-Touch Based 3D Interaction Method
Jinki Jung, Jihye Hong, Sungheon Park, Hyun S. Yang
VRCAI, 2012
paper

Template code of this website was stolen from here.