[zmmmmm]

The shocking thing that at least I ran into is the sheer quantity of training data you really need. The large companies doing this successfully are using utterly gigantic libraries of training data that are beyond anything others could ever come up with. It really brought home to me what a blunt intstrument deep learning really is.

[fareesh]

Is there some kind of rule of thumb for a minimum of how much data is needed for various types of problems?

[TACIXAT]

Retraining and existing model does not need many (fast.ai lesson 1 example is retraining a net to distinguish cricketers and baseball players with 30 images). For a full net, it's on the order of thousands per category.