Road Sign Detection Model
Custom YOLOv7 object detection trained on Finnish road signs
Project Overview
I built this computer-vision pipeline from scratch for my Bachelor thesis at SAMK, including data collection, manual annotation, training, and evaluation.
The dataset was captured in real Finnish conditions with weather and lighting variation, then labeled class-by-class using LabelImg. The final YOLOv7 model reached around 80% mAP@0.5 and performed best on medium and large signs under realistic road scenes.
This project proved my ability to execute a full ML lifecycle, from raw data to model analysis and thesis documentation.
Key Highlights
- Self-collected & annotated dataset
- ~80% detection accuracy
- Real-world condition testing (weather, lighting, occlusion)
- Precision & recall evaluation
- Custom dataset pipeline
- Model fine-tuning & hyperparameter optimization
Related Projects
2026
TyöHunt
Full-stack SaaS that scrapes Finnish job boards, scores every listing against your CV with Claude AI, and generates tailored cover letters.
Open case study2025
Finnish Job Scraper
The v4 full-stack platform that combined FastAPI, Claude AI matching, and multi-source scraping — later evolved into TyöHunt SaaS.
Open case study