Figure out what ML library we should use in the future. (for stuff like YOLO)

[Mysticial]

We can't move into real ML without a library.

@denvoros has decided we will use pytorch for training. But we still need the C++ inference side.

[pifopi]

done ?

[Mysticial]

OpenCV looks like it'll work, but probably a bit too early to make any calls yet. Will be easier to decide when we get actual ML stuff into production.