I really wish they would open source some of the system or at least write papers outlining the algorithms used. Also I wonder if we can have a neural net that

I really wish they would open source some of the system or at least write papers outlining the algorithms used. Also I wonder if we can have a neural net that generates these classical approaches given a class of objects to identify. Or maybe once you have trained a neural net to work on recognising basic features, you could transform (compile?) it to an algorithm that can be debugged and expanded.