I'm doing a project about doing OCR on receipts and today, while trying to do
character segmentation, I made a pretty stupid mistake that led to my model
overfitting almost perfectly pretty neatly (in some cases I got 100% correct
classification
