Training data used in the Core ML feature of Talk Dim Sum. Currently trying to balance with about ten images per category.