AI-generated project mascot visual for GitHub branding. Not a personal portrait.
AI/ML · Computer Vision · Road Damage Segmentation · Forecasting · LLM Applications
Computer Science undergraduate at Hankuk University of Foreign Studies.
I build research-oriented AI/ML projects that connect model design, experiment evidence, visual analysis, and deployable portfolio presentation.
한국외대 컴퓨터공학전공 4학년 재학중입니다. AI/ML·컴퓨터비전 중심으로 도로 세그멘테이션, 객체 탐지, 금융 예측, LLM 응용 시스템등등 설계,개발합니다.
- University: Hankuk University of Foreign Studies (한국외국어대학교)
- Major: Computer Engineering (컴퓨터공학전공)
- Double Major: Northeast Asian Diplomacy and Commerce (동북아시아 외교통상전공)
- Languages: Korean · English · Japanese (한국어, 영어, 일본어)
| Area | Description |
|---|---|
| Computer Vision | Semantic segmentation, object detection, visual evidence analysis |
| Road Damage Segmentation | Boundary representation degradation, lightweight segmentation, boundary-guided fusion |
| AI / Data Forecasting | Financial prediction, urban scenario forecasting, structured-data modeling |
| LLM Applications | Prototype-level LLM service design and AI-assisted reporting |
| Deployment & Portfolio | GitHub Pages, static demos, experiment portals, research documentation |
LiteRaceSegNet is a lightweight semantic segmentation project for road-damage analysis.
The project investigates boundary representation degradation in road-damage segmentation and explores a lightweight Boundary-Guided Fusion architecture for improving boundary preservation.
Core keywords
Semantic Segmentation · Boundary-Guided Fusion · Lightweight CNN · Road Damage Analysis · CPU/GPU Evidence · Ablation Study
Repositories
| Repository | Purpose |
|---|---|
| LiteRaceSegNet-V13-Portal-Clean | Research portal, visual evidence, presentation-ready project archive |
| LiteRaceSegNet-V11 | Experimental implementation and boundary-aware segmentation pipeline |
TRACE-Eval refers to my prototype evaluation framework for low-precision inference stability and deployment readiness.
In this portfolio, TRACE-Eval is used for checking FP32 / FP16 / INT8 behavior, runtime consistency, latency changes, and compatibility issues across inference settings.
It is an independent academic/portfolio project name and is not affiliated with any similarly named external products, services, or organizations.
HPLS-Eval is kept separate for medical / lesion-related evaluation experiments.
I separate TRACE-Eval and HPLS-Eval to avoid mixing road-damage deployment evaluation with medical-domain stability experiments.
| Project | Description |
|---|---|
| Financial Future Prediction Service | Machine learning project for financial and company-performance prediction |
| Machilens: Shibuya AI Predictor | Urban scenario forecasting demo focused on Shibuya |
| LEO Handover AI Optimizer | AI-assisted LEO satellite handover optimization prototype |
AI / ML
Web / Deployment
My portfolio focuses on building AI systems with visible evidence:
- model design
- dataset and experiment organization
- quantitative evaluation
- visual comparison
- research-style documentation
- deployable demo pages
Avatar: AI-generated.
This GitHub profile is maintained as an academic and portfolio-oriented project archive.
Clean GitHub Pages portfolio for a lightweight road-damage semantic segmentation experiment with boundary-guided analysis, CPU/GPU evidence, and visual result documentation.