Well, it uses OpenCV and Python currently, which are hardly easy to set up. So it'll either be a webapp (do the processing server-side), or I'll need to reimplement it or repackage it to make it a simple, cross-platform install.
The theory is quite similar to a QR Code scanner, looking for salient aspects of an image and deriving data from that.
For the photo process, there could probably be multiple vectors, drag onto page, browse, open camera, etc... The API should be made in such a way that other frontends could be built, like native apps or an interface directly in Unity.
For the different types of level, I'm more worried about the representation rather than the game genre. Right now it only knows about lines, which can be used to describe a system of walls, floors or outlines. It'd be nice to be able to tag the polygons in between the lines through colour-coding (here is lava, for instance). A tile-based representation could be cool too (need to drop by the shop and buy some felt-tip pens to test this)
Would you think the system would just interpret the scan in terms of simple line objects, then let the user tweak each line accordingly? I wonder if it would speed up the process, and make it seem a bit more magic, if you could say "I want to make a platformer," and then all the lines are interpreted as floors. If you say "Let's build a dungeon crawler," it changes to top-down mode and lines are all walls. In fact, for a simple proof of concept, that's probably as much as I'd recommend building right there, and adding more functionality to it later. Still, it's worth throwing all sorts of ideas out there.
