Dataset details: https://universe.roboflow.com/datasets-84fcz/airplane-vgvsf/dataset/1
Dataset description
Dataset contains 4140 images of airplanes. For training dataset we'll take 2898 (70%) images and validate on 828 (20%).
Model
For object detection we will use yolov8s model from Ultralytics
Detection example
Confusion matrix
Loss and metrics
Planned improvments
As the graphs shows we already have pretty nice results - the model converges and the training might me continued. But there are steel techniques that could be implementec for achieving better results.
- Use more data augmentation to avoid overfitting
- Use trained model for more complex task of counting airplanes on the airfield (bird's eye view shots)