Great explanation. The fastest way to do this I know of is to do the first hour of the fast.ai deep learning course, replacing the cats and dogs images with 1000 square hotdog pics and similar volume of other food via Google images search. I created a wrapper to create test and validation sets automatically. Fast.ai has a bunch of routines for data augmentation built in too.

Or just use lobe.ai and not write any code at all.

    Julian Climate Crisis Harris

    Written by

    Technical Product guy specialising in the impact of machine learning on speech and text (NLP). Ex-Google, but none of this matters if we don't fix the planet.