This is a bit different... this is not about a single photo, but a way of presenting a scene of them. Feed it a load of photos of the same scene (place, object, whatever) and it'll calculate where each one was taken and from there where others are relative, such that you can kinda navigate between them.
I had a quick play with some old pics, and click for the interior of the Colosseum in Rome
This also reminds me of something I saw before, but can't remember where or what it was called. That took a single image, and using the visual information, reconstructed a limited 3D scene from the single 2D photo. You could then move about near the camera spot, and the perspective would change accordingly.
Now we just have to wait for them to put the two together and generate walkaround 3D scenes from an arbitrary bucket of photos.