It’s not surprising that Sony’s depth sensor model of Alban has even less 3D information than the 2D photogrammetry model, even though the depth sensor acquires 3D geometry directly. The Sony 3D sensor isn’t very good at extracting that third dimension, though the depth sensor data would be reconstructed into the 3D mesh faster than the 2D only photogrammetry.

Overall, however, I disagree with your comment. While it would be swell for 3D to magically pop out from 2D only camera(s), it’s not going to happen. There will need to be an active depth sensor working with 2D imaging sensors to pull off AR/VR/3D/ etc… it’s just the way it is. Machine learning won’t break this constraint, that’s my prediction.