jseobyun.github.io

About me

Hello, I’m Jongseob Yun. Thanks to visit this page! I’ve focused on computer vision and computer graphics that handle human stuffs(face, body pose/shape) and 3D geometric tasks (SfM, MVS, surface reconstruction, neural rendering). I am currently working at NAVERLABS, Korea. Please contact me if you need through e-mail!

Curriculum Vitae : link
LinkedIn : link
Blog (Korean only) : link
Contact : jseob.y@naverlabs.com

Education

1) Master degree from VILAB advised by Prof. Kuk-Jin Yoon, Korea Advanced Institute of Science and Technology(KAIST)
2) Baschelor’s degree at Pohang University of Science and Technology(POSTECH)

Expertise

1) Camera intrinsic/extrinsic calibration(both pinhole, fisheye) and Image warping
2) Optimization based on CERES (e.g. 6DOF pose estimation)
3) Neural rendering / surface reconstruction
4) 2D/3D Human pose and shape estimation
5) Omnidirectional image handling
6) Structure from motion and Multiview Stereo, Feature matching

Skills

1) Python (numpy, torch, opencv, etc.) 👍👍👍👍👍
2) C++ (boost, tbb, opencv, ceres, etc.) 👍👍👍
3) Docker/Git/Linux 👍👍👍
4) Calibration tools (camodocal, basalt) 👍👍👍

Publication (Conference)

SpherePHD: Applying CNNs on a Spherical PolyHeDron Representation of 360◦ Images
Yeonkun Lee*, Jaeseok Jeong∗, Jongseob Yun∗, Wonjune Cho∗, Kuk-Jin Yoon
published in CVPR2019, co-1st author
[PAPER]

Publication (Journal)

SpherePHD: Applying CNNs on 360∘∘ Images With Non-Euclidean Spherical PolyHeDron Representation
Yeonkun Lee*, Jaeseok Jeong∗, Jongseob Yun∗, Wonjune Cho∗, Kuk-Jin Yoon
published in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI 2020), co-2nd author
[PAPER]

Experience

Reproducing and Extending OpenPIFPAF

I reproduced 2D HPE algorithm, OpenPIFPAF and extended it by adding bounding box module and absolute human distance module.

Making 3D human pose annotation tools

I made 3D human annotaion tool(GUI) based on epipolar geometry.

Fitting SMPL to RGBD images

I made optimization based SMPL fitting algorithm on RGBD images. The pipeline follows SMPLify and VideoAvatar.

Rerpoducing EverybodyDanceNow (2D motion synthesis) for fun

For practicing GAN framework, I reproduced EverybodyDanceNow pytorch ver. based on arxiv paper.

Reproducing KinectFusion and VisualHull

I reproduced KinectFusion and VisualHull pytorch GPU ver.

Following up 3D neural surface reconstruction SOTAs

I've followed up NeRF series that focusing on reconstructing surfaces. Additionally, I invented the method to fill out unscanned region of rigid objects.

Various camera model reproduction and calibration

I've covered various camera model(BC, KB, UCM, DS, etc.) and reproduced them using numpy and pytorch, so I'm familiar with undistortion and warping among camera models. Additionally I've implemented extrinsic calibration pipepline based on CERES by myself.

Career

NAVERLABS, KOREA (2020.12 ~ now)

I've been working at Robot vision team in Vision group as Technical Research Personnemilitary which is one kind of military service.

KETI, Pangyo (2020.8 ~ 2020.12)

I shortly worked human team as an internship.

KAIST, Daejeon (2018.8 ~ 2020.8)

I graduated VILAB advised by Kuk-Jin Yoon after finishing Master degree course.

POSTECH, Pohang (2013.3 ~ 2018.8)

I graduated Mechnical enginerring department with 1 year leave of absence. During the leave of absence, I worked at start-up company that developed Robot control system.

Honor

The 31st Image Processing and Image Understanding(IPIU) Best paper award Winner
Our paper mentioned as Best paper of the 31st Image Processing and Image Understanding(IPIU)

On Dream Future Industry Graduate School Scholarship
I was selected the student of On Dream Future Industry Graduate School Scholarship 2019-2020

Personal Things

Basketball Team, POBBA (POSTECH)
I've been basketball team member since 2014 and still enjoyed games with our sernios and juniors. Additionally I've done steadily workout for 8 years.

</p>