About

I work on generative models for image restoration and generation, which I treat as inverse problems. I am mainly interested in reducing their need for task-specific training and data, using training-free and few-step methods that reuse the priors already in pretrained models. These apply across object removal and inpainting, few-step image and video generation, and reconstruction in both natural and medical imaging.

I am currently an Integrated M.S. & Ph.D. student at Seoul National University, and I hold a dual B.S. in Biomedical Engineering and Artificial Intelligence from Korea University. I also research at OGQ.

Research Interests
Diffusion & Flow Matching Object Removal & Inpainting Few-Step Generation Training-Free & Test-Time Methods Inverse Problems Video Generation Medical Imaging (MRI / CT / X-ray)
News
2026.06
EraseLoRA accepted to ECCV 2026.
2026.05
Secured a MOTIR Industrial Technology Development grant on multimodal AI for intelligent design as lead proposal author.
2026.02
GH-NAF accepted to CVPR 2026.
2025.06
OFF-CLIP accepted to MICCAI 2025.
Publications
EraseLoRA
ERASELORA
ECCV 2026
Sanghyun Jo*†, Donghwan Lee*, Eunji Jung*, Seong Je Oh, Kyungsu Kim
European Conference on Computer Vision (ECCV), 2026 · Co-First Author
Dataset-free object removal: MLLM-driven foreground exclusion paired with background subtype aggregation.
GH-NAF
GH-NAF
CVPR 2026
Seongje Oh*, Ju Hwan Lee*, Chae Yeon Lim*, Donghwan Lee, Myungjin Chung, Kyungsu Kim
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
Decouples hash levels and adaptively weights them via uncertainty-guided, grid-adaptive hash-level attention in NeRF.
OFF-CLIP
OFF-CLIP
MICCAI 2025 · Early Acceptance (Top 9%)
Junhyun Park*, Chanyu Moon*, Donghwan Lee, Kyungsu Kim, Minho Hwang
Int. Conf. on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2025
Raises normal-case detection confidence in radiology CLIP via an off-diagonal term auto-adjustment.
Curriculum Vitae

Education

2025.03 – Present

Integrated M.S. & Ph.D., Biomedical Sciences

Seoul National University

Advisor: Prof. Kyungsu Kim

2021.03 – 2025.02

B.S., Biomedical Engineering & Artificial Intelligence (dual)

Korea University

GPA 3.9 / 4.5

Experience

2024.11 – Present

AI Research Engineer

OGQ

SOTA generative AI with diffusion and flow matching — image/video inpainting, object removal, and editing/composition.

2025.04

Reviewer

ICCV 2025

Reviewed computer vision submissions with Prof. Kyungsu Kim.

2025.04

Research Proposal

SNU R&DB Foundation

Core-AI mapping of real-time EEG dynamics to MRI-standard biomarkers for brain-tumor monitoring and prognosis.

2023.12 – 2024.08

Research Intern

Magnetic Resonance Lab, SNU

Under-sampled MRI reconstruction via cross-domain CNNs with data consistency.

Grants & Funding

2026.05 – 2028.12

Development of Multimodal Data Integration Technology and an AI Service Platform for Intelligent Design Process Management

Ministry of Trade, Industry and Resources (MOTIR)

Lead proposal author; wrote the winning proposal and secured the grant (PI: Prof. Kyungsu Kim). ₩600M (≈ US$390K) over three years.

Awards

2025.02

Academic Excellence Award (2024)

Korea University
2024.02

Research Excellence Award, Undergraduate Research Program

Dept. of Biomedical Sciences, SNU College of Medicine