@jongold thanks! This whole project actually just started as a curiosity about whether something like VGG16 could do well in a domain it wasn't intended to do - and it did surprisingly well given the right dataset. I loved your article and the line of thinking from programmatic design. It's really interesting to see how 2 years made such a difference, now with things like VGG16, we can assume that machines already can see, and direct that ability towards new domains in service of new tools.

@joid thanks for the heads up! should be fixed now.