I’m sure this is largely possible by composing existing techniques, but still, that’s some kind of wonderful.  The computer is basically understanding what it sees and generating new views on the fly.  That’s…yeah.

Seen at BoingBoing