Education
-
2023.09 - Present Gwangju, South Korea
Integrated M.S./Ph.D. — Gwangju Institute of Science and Technology (GIST)
AI Graduate School
-
2018.03 - 2023.09 Wonju, South Korea
B.S. — Yonsei University (Wonju)
Computer Engineering
Work
- 2023.09 - Present
Integrated M.S./Ph.D. Student — GIST Autonomous Driving Lab
Generative AI / Autonomous Driving Research
- 2023.01 - 2023.03
AI Intern — InBic AI
AI edge device (Jetson) environment setup and operation experiments
Projects
- 2025.07 - Present
바이오 체화형 피지컬 AI 연구단
Bio-Embodied Physical AI Research Center
- Diffusion-based future prediction model development (디퓨전 기반 미래예측 모델 개발)
- 2024.09 - 2025.12
Self-Photography Pose Recommendation System Using Pose Estimation and Object Recognition
- Full-stack web system development (React, Django, MariaDB)
- AI model web-based deployment and real-time service
- 2024.09 - 2025.12
Low-Light Image Object Detection
- Low-light image data collection and labeling
- Object detection model development and fine-tuning
- On-device demonstration on Qualcomm RB5 board
- 2024.06 - 2024.12
실도로 위험상황 시나리오 기반 시뮬레이션 데이터
Simulation Data Based on Real-Road Hazard Scenarios
- Vision-Language Model development and fine-tuning (VLM 개발 및 파인튜닝)
- 2024.05 - 2026.04
다종센서 정보를 활용한 딥러닝 기반 표적 탐지 및 추적기술 개발
Deep Learning-Based Target Detection and Tracking Using Multi-Sensor Information
- Hyperspectral data collection and analysis (초분광 데이터 수집 및 분석)
- Hyperspectral detection and generative model development (초분광 탐지 및 생성 모델 개발)
- 2024.01 - 2027.04
미래형자동차핵심기술전문인력양성
Future Automotive Core Technology Professional Training Program
Publications
-
2026.02 TDiff-HSI: Tucker-guided diffusion for high-dimensional RGB-to-HSI image generation
Journal of Computational Design and Engineering, Oxford University Press
Jaeik Bae, Yong-Gu Lee
We introduce TDiff-HSI, a diffusion-based model that can generate hyperspectral images (HSIs) directly from RGB images and material-wise segmentation masks. HSI provides both spatial (u, v) and spectral (λ) information. The accompanying dataset that we are releasing spans wavelengths in the range from 420 to 1728 nm, digitized into 512 channels. Directly handling this immense three-dimensional dataset is computationally prohibitive and often leads to numerical errors. To address this challenge, TDiff-HSI leverages Tucker decomposition to reduce dimensionality, enabling more stable and efficient processing. Moreover, spectral precision is enhanced by combining RGB channels with a material segmentation mask. To support this research, we constructed a new dataset using a hyperspectral camera. The dataset comprises 40,014 RGB-HSI pairs across 78 scenes, featuring 12 objects with diverse material properties.
-
2025.11 BISeg: Band-Interval Selection for Explainable Hyperspectral Segmentation
CIKM 2025 Workshop on Human-Centric AI (HCAI), COEX, Seoul, Korea
Jaeik Bae, Yong-Gu Lee
We introduce BISeg, a novel hyperspectral image (HSI) segmentation model that achieves both accuracy and interpretability through band-interval selection. Unlike conventional approaches that rely on post-hoc band attribution, BISeg intrinsically learns to identify compact spectral windows during training. Each interval is parameterized by its center wavelength and width, with a maximum of ten intervals activated. A dedicated regularization encourages the use of fewer and narrower intervals, ensuring explanations remain concise and physically meaningful. The proposed spectral–spatial backbone integrates these intervals with spatial features via cross-axis attention, and a query-based head outputs pixel-level mask. Experiments on our custom HSI dataset, covering nine material classes, demonstrate that BISeg achieves competitive segmentation accuracy while producing interpretable spectral intervals aligned with known material properties.
-
2025.06 R2H-Diff: Spectral-wise Diffusion for Hyperspectral Image Reconstruction from RGB and Semantic Masks
Korean Society for Military Science and Technology Annual Conference 2025
Jaeik Bae, Yong-Gu Lee
We propose R2H-Diff, a diffusion-based model that reconstructs hyperspectral images (HSIs) using RGB images and semantic masks. Instead of generating the full spectrum at once, our model sequentially restores each spectral channel for greater training stability. A single-scene experiment validates its ability to denoise and reconstruct spectral data effectively.