Data augmentation as a method for model improvement

[epierotti3]

Dependencies

• (insert as needed)

Overview

We'd like to augment (e.g. rotate) at least some of the training data images as a way to make more training data images instead of creating entirely brand new images to train the model on. This should be implemented in tensor flow and should be pretty easy to incorporate to the existing project.

Action Items

• Research augmentation methods that are compatible with our current images and model
• Determine where in "the code" this needs to go
• "Test" on a few images
• If works, apply to more images and train the model on this to see if accuracy improves.

Resources/Instructions

(add as needed)