This repo is an attempt at using object detection from VODs provided by The Blue Alliance to gain statistical insights beyond the conventional FIRST API.
The dataset can be found here. The current dataset focuses on FIM competitions in particular for consistency in camera conditions. Could use more diversity in the dataset for better robot detection.
Programs require you to input absolute paths at the moment. Currently only supports running on Ubuntu, likely not too hard to hack it to work on other operating systems though.
I recommend looking over the Yolov8 docs for more info on how training and implementation of the neural network is handled.
NOTE: The current requirements.txt is broken.