Summary:

Transfer learning was applied to a pretrained YOLOv5 model for swimming pool detection on satellite image detection. The training and validation data was obtained from google satellite basemap, which was exported using ArcMap. The large exported image was tiled to 640*640 size tiles, which were then labelled using Roboflow. These labelled images were then used to train the YOLOv5 model for swimming pool detection. The model was then deployed as a streamlit web application.