Finding relevant ROI's and creating bounding boxes around them.
Instead of training a model with the images without any preprocessing, I wanted to see if creating ROI's by thresholding and then training a model on them would help increase the accuracy.