Dong-Hyun Hwang, Ph.D.
Research Interests
- Computer Vision and Machine Learning
- Multi-modal Generative Models (Text-to-Image, Video, Motion, etc.)
- Large-scale Model Training and Engineering
- Human Pose Estimation
- Human-Computer Interaction
- Interactive Systems using Computer Vision
- System Usability Evaluation and Statistical Analysis
Professional Experience
Manager, Noetra株式会社
07/2026 - Present
Computer Vision Lead Research Engineer, SB Intuitions
08/2025 - Present
- Research and development of image-generation models.
Research Engineer, NAVER Cloud
01/2023 - 06/2025
- Team Lead of Hyperscale AI Model101 (2024 - 2025)
- Led a team of 4 to develop an AI Eraser Agent in CLOVA X, successfully launching the service.
- Contributed to a large-scale diffusion-based text-to-image training framework.
- Team Lead of Avatar (2023)
- Directed a team of 6 in virtual human agent technology development.
- Recognized with the N INNOVATION Award for Outstanding Technology in R&D.
Research Engineer, CLOVA, NAVER
02/2022 - 12/2022
- Built a multi-view motion capture studio for high-precision 3D motion data.
- Researched motion generation using generative AI models, focusing on motion augmentation and text-to-motion synthesis.
Other Experiences
- JSPS Research Fellowship (DC2) (04/2020 - 03/2022)
- Research Assistant, Team Koike, JST (12/2018 - 03/2022)
- Program Chair, ACM CHI2021 Workshop (04/2021)
- Visiting Scholar, Kris Kitani Lab, Carnegie Mellon University (10/2019 - 01/2020)
Education
Tokyo Institute of Technology, Tokyo, Japan
- Ph.D. in Computer Science (04/2019 - 03/2022)
- Thesis: Markerless Human Motion Capture and Visualization from Monocular Videos
- M.S. in Computer Science (04/2017 - 03/2019)
- Thesis: Synthesizing Pseudo-2.5D Content from Monocular Videos for Mixed Reality
Korea University of Technology and Education, Korea
- B.Sc. in Computer Engineering (03/2010 - 02/2017)
- Thesis: A Pointing Device Using Hand Gesture Recognition (summa cum laude)
Skills
- Programming Languages: Python, C++
- Machine Learning Frameworks: PyTorch, TensorFlow
- Tools: OpenCV, Unity, FastAPI, Gradio
- Distributed Training: PyTorch DDP, DeepSpeed
- Languages: Korean (Native), Japanese (Fluent - JLPT N1), English (Advanced)
Publications
- ZIM: Zero-Shot Image Matting for Anything (ICCV 2025, Highlight) — Paper
- [1] Diffusion-based Synthetic Dataset Generation for Egocentric 3D Human Pose Estimation (ECCV 2024 Workshop)
- [3] MonoEye: Multimodal Human Motion Capture System Using A Single Ultra-Wide Fisheye Camera (UIST 2020)
- [4] Lightweight 3D Human Pose Estimation Network Training Using Teacher-Student Learning (IEEE/CVF WACV 2020)
View all publications
Awards & Grants
- N INNOVATION Award, NAVER Corporation
- Grant-in-Aid for JSPS Fellows (KAKENHI Project)
Patents
- Motion Measurement Device (JP2022007839A)
- Method for Generating Data for 3D Pose Estimation (KR102562378B1, JP7178396B2, US11610331B2)