All projects

Project 2 of 3

Team project

Defect detection on printed circuit boards

YOLOv8 models fine-tuned to detect surface defects, then measured for missed defects and false alarms.

Measured on a separate test set

Missed defects False alarms

Illustration of a circuit board with two detection boxes drawn around defects.
Period
Dec 2025 – Mar 2026
Type
Team project
Course
Advanced Machine Learning for Computer Vision
University
IT University of Copenhagen
Tools
  • YOLOv8
  • Weights & Biases
Code
GitHub repository

Method

  1. Fine-tuned YOLOv8 models to detect surface defects
  2. Kept training runs repeatable with configuration files
  3. Tracked detection metrics in Weights & Biases
  4. Compared saved checkpoints and chose a threshold on validation images
  5. Measured missed defects and false alarms on a separate test set