This is absolutely mind-blowing! I'm looking forward to reading the paper they published.
This may be obvious to some, but the software already knows what's on the colouring book page. It has a predefined model of the character, so it just applies a texture mapping algorithm (with some guessing to colour the back side of the character).
I actually just started reading the abstract and let out a deep sigh. It's still impressive but no longer mind-blowing unfortunately.