This. Google has so much data at this point they should be actively converting it all to pure 3D (structure from motion etc). With tango on the horizon there will be a point in time phone could know where its at by simply comparing geometry, or at the very least augment camera view with accurate directions based on the scene geometry, and not crude 'somewhere this way' gps+magnetometer Arrow. They could also use this data to augment self driving cars, so its a no-brainer.

Google Earth is PATHETIC right now, sparse hand made badly textured blocky models of few selected buildings in a capital of European country? Really Google?

I sure hope they have a group inside google working on this already.