Video Understanding
Localizing abnormal events with weak supervision.
Junhee Lee
AI Researcher
I am an AI researcher interested in multimodal and video understanding, representation learning, and open-vocabulary perception. My research focuses on building robust and generalizable visual intelligence systems.
Research Interests
AI-Generated Media Forensics
Localizing AI-generated and edited image regions.
3D Scene Understanding
Instance-level perception from 3D and multi-view observations.
Label-Efficient Vision Learning
Learning visual representations with limited annotations.
About
Introduce yourself here: your current position, affiliation, research interests, and what motivates your work. Keep this paragraph personal and specific to your own research.
Yongin, Korea
Selected Publications
All publications
Less Supervision, Better Generalization: Weakly Supervised Fake Region Localization in Diffusion-Edited Images
NeurIPS, 2026
RefineVAD: Semantic-Guided Feature Recalibration for Weakly Supervised Video Anomaly Detection
AAAI, 2026
Projects
3D MRA Vessel Mapping for Robot-Assisted Cerebrovascular Intervention
Segmenting and enhancing cerebral vessels in 3D MRA to construct informative vascular maps for optimizing robot-assisted cerebral aneurysm intervention paths.
- 3D Medical Image Segmentation
- 3D MRA
- Robot-Assisted Intervention
Thermal–RGB Fusion for Robust Nighttime Anomaly Detection
Developing thermal–RGB fusion to improve anomaly detection robustness in nighttime and low-visibility environments.
- Video Anomaly Detection
- Thermal–RGB Fusion
- Low-Light Vision
Experience
Jun. 2024 – Feb. 2026
Undergraduate Research Intern
Visual Science Lab (VSLab), Kyung Hee University
Education
Mar. 2026 – Present
M.S. in Computer Engineering
Kyung Hee University
Mar. 2022 – Feb. 2026
B.S. in Artificial Intelligence
Kyung Hee University
GPA: 3.94 / 4.3 (4.2 / 4.5)
Contact
I’m open to research collaborations, academic discussions, and new opportunities. Feel free to reach out by email or connect through the links below.