Portfolio
비공개 포트폴리오입니다.
direcf.github.io / personal / portfolio
Seungjun Lee
AI / Vision Research Engineer
Summary
AI Research Engineer with a strong background in computer vision and multimodal AI, specializing in first-person (egocentric) video understanding and real-time event analysis systems. Experienced in building end-to-end AI pipelines from research to production deployment. Holder of 9 domestic patents and recipient of 3rd Prize at CVPR 2024 AVA Keypoint Challenge. Currently pursuing research at the intersection of egocentric and exocentric vision (Ego-Exo) with the goal of contributing to the academic community.
Education
Korea Advanced Institute of Science and Technology (KAIST) 2023 – Present
M.S. in Electrical Engineering · Vision AI Lab
- Research focus: Egocentric video understanding, Ego-Exo cross-view learning, multimodal AI
- GPA: TBD
Hanyang University 2017 – 2023
B.S. in Convergence Electronics Engineering
- GPA: TBD
Technical Skills
Languages & Frameworks
Python PyTorch C++ CUDA FastAPI TypeScript
Research Areas
Egocentric Video Understanding Ego-Exo Cross-View Learning Action Recognition Object Correspondence Real-time Video Analysis VLM / Multimodal AI
Infrastructure
AWS Docker Kubernetes NVIDIA Triton Redis Kafka
Publications
In preparation — targeting CVPR / ICCV 2027.
Patents
9 domestic patents registered (Korea Patent Office)
실시간 이상행동 감지 시스템 1인칭 영상 기반 객체 추적 방법 다중 카메라 동기화 프레임워크 VLM 기반 안전 모니터링 방법 발달장애 행동 분석 AI 모델 엣지 디바이스 경량 추론 방법 이벤트 기반 비디오 압축 방법 크로스뷰 객체 대응 학습 방법 산업 현장 위험 예측 시스템
Awards & Challenges
CVPR 2024 — AVA Keypoint Challenge Jun 2024
3rd Prize (Bronze) · ActivityNet Workshop, CVPR 2024
- Developed a multi-scale human pose estimation pipeline on the AVA dataset
- Achieved 3rd place globally among participating teams
Key Projects
Carevia
SK Telecom · Developmental Disability Support
AI-powered behavioral analysis system for individuals with developmental disabilities. Leveraged egocentric video and multimodal signals to detect behavioral patterns, enabling real-time support alerts for caregivers.
Egocentric Vision Behavior Analysis EdgeAI
SynapseGo
1st-Person VLM Video Analysis
End-to-end first-person video understanding platform using Vision-Language Models. Supports real-time scene narration, object tracking, and action recognition from wearable camera streams.
VLM Egocentric Real-time
Industrial Safety System
Physical Security · Real-time Event Analysis
Real-time event detection and analysis platform for industrial safety and physical security. Multi-camera ingestion with sub-second anomaly alerting, deployed at scale across manufacturing sites.
Multi-Camera Event Detection Streaming
Ego-Exo Research
KAIST Vision AI Lab · Ongoing
Research on cross-view correspondence and transfer learning between egocentric and exocentric video streams. Targeting CVPR-level publication on temporal action-level cycle-consistent alignment.
Ego-Exo Self-supervised Research
Last updated · July 2026 · Details available upon request