Byungki Kwon

I am a Ph.D. student at POSTECH, South Korea, advised by Prof. Tae-Hyun Oh.
My research focuses on generative models for images, videos, and multi-modal data (e.g., RGB–depth), with a particular emphasis on diffusion-based architectures. I am interested in enhancing user-perceived quality and practical efficiency of generative systems by incorporating geometry- and motion-aware modeling.
During my M.S. studies at POSTECH, I worked on motion analysis and depth estimation, which provided a strong foundation in modeling scene structure and dynamics.

Keywords: generative model / motion estimation / depth estimation / but not limited to

Contact: byungki.kwon[at]postech[dot]ac[dot]kr

Information: CV, Google Scholar, Linkedin

Work Experience

Research Internship, Sony AI

Apr. 2026 – Present

Research Internship, Naver Cloud

Oct. 2025 – Apr. 2026

Research Internship, Microsoft Research Asia

Sep. 2024 – Mar. 2025

Education

Ph.D. of Artificial Intelligence (advisor: Tae-Hyun Oh)
Master of Electrical Engineering (advisor: Tae-Hyun Oh)
Bachelor of Mechanical Engineering

POSTECH, Mar. 2023
POSTECH, Sep. 2020 – Feb. 2023
POSTECH, Mar. 2013 – Aug 2019

Publications

Early Failure Detection and Intervention in Video Diffusion Models, NeurIPS 2026

Byungki Kwon, Sohwi Lim, Nam Hyeon-Woo, Moon Ye-Bin,
Tae-Hyun Oh

Zero-Shot Rankability: Revealing Latent Ordinal Structure in Multimodal Large Language Models via Language, ICML 2026

Nam Hyeon-Woo, Moon Ye-Bin, Sohwi Lim, Byungki Kwon, and Tae-Hyun Oh

HDR-NSFF: High Dynamic Range Neural Scene Flow Fields, ICLR 2026

Shin Dong-Yeon, Kim Jun-Seong, Byungki Kwon, Tae-Hyun Oh

HDR-NSFF.mp4

Revisiting Deep Video Motion Magnification for Real-time Applications, TMLR 2026

Ha Hyunwoo, Oh Hyun-Bin, Kim Jun-Seong, Byungki Kwon, Kim Sung-Bin, Ji-Yun Kim, Sung-Ho Bae, Tae-Hyun Oh

realtime_mm.mp4

JointDiT: Enhancing RGB-Depth Joint Modeling with Diffusion Transformers, ICCV 2025
*Work performed while interning at Microsoft Research Asia (MSRA)

Byungki Kwon, Qi Dai, Lee Hyoseok, Chong Luo, Tae-Hyun Oh

jointdit.mp4

Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior, AAAI 2025

Lee Hyoseok, Kyeong Seon Kim, Byungki Kwon, Tae-Hyun Oh

Learning-based Axial Video Motion Magnification, ECCV 2024

Byungki Kwon, Oh Hyun-Bin, Kim Jun-Seong, Ha Hyunwoo, Tae-Hyun Oh

axialmm.mp4

Uni-DVPS: Unified Model for Depth-Aware Video Panoptic Segmentation, RA-L 2024

Kim Ji-Yeon, Oh Hyun-Bin, Byungki Kwon, Dahun Kim, Yongjin Kwon, Tae-Hyun Oh

The Devil is in the Details: Simple Remedies for Image-to-LiDAR Representation Learning, ACCV 2024

Jo Won-jun, Byungki Kwon, Kim Ji-Yeon, Hawook Jeong, Kyungdon Joo, Tae-Hyun Oh

DFlow: Learning to Synthesize Better Optical Flow datasets via a Differentiable Pipeline, ICLR 2023

Byungki Kwon, Nam Hyeon-Woo, Kim Ji-Yun, Tae-Hyun Oh

The Devil in the Details: Simple and Effective Optical Flow Data Generation, TVCJ 2024

Byungki Kwon, Kim Sung-Bin, Tae-Hyun Oh

Awards

Outstanding Reviewer Award, ICCV 2023
Samsung HumanTech Paper Award, 2023 (5,000$)
Best Paper Award, Asian Federation of Computer VIsion (AFCV), 2024
Best Paper Award, Workshop on Image Processing and Image Understanding (IPIU), 2023
Best Paper Award, Workshop on Image Processing and Image Understanding (IPIU), 2021

Page updated

Google Sites

Report abuse