Oh! Ack! Thanks! Very good question. I'll try to remember to put a 'learning comment' for that on the next go-around. Those are the means and standard deviation

Oh! Ack! Thanks! Very good question. I'll try to remember to put a 'learning comment' for that on the next go-around. Those are the means and standard deviations of each channel in the training set -- normalizing them up front helps us expect that we have a 0 mean and variance of 1 with our images going into our neural network. This is helpful as we don't need a BatchNorm in front of our first layer that way (and I'm sure for other reasons too). Many thanks for asking, that's a very great question! I was super scared for a minute that I somehow had a naked hyperparameter untagged, but these certainly are very ~mysterious~ numbers indeedy!~ :D <3 <3 <3 <3 :D :D :D :)))