YOLO

YOLO

YOLO branch of the repository contains all the assets and components required to train a YOLOv3 real-time object-detection system that the AUV would use to detect objects underwater.

For our dataset, we mixed BeaverAUV's open sourced dataset that contains around 1700 images of relevant underwater objects with our own. We have also extracted frames from Youtube videos of AUV competitions for a more diverse dataset.

The Pascal-VOC based image annotations provided by the BeaverAUV dataset had to be converted to YOLO's format (as shown below) using python scripts that parse XML annotations and map the corresponding values.

[category number] [object center in X] [object center in Y] [object width in X] [object width in Y]