I'm interested in what Apple have cooking, but stereo + depthmap doesn't seem like enough information to accurately construct a perspective correct scene outsid

I'm interested in what Apple have cooking, but stereo + depthmap doesn't seem like enough information to accurately construct a perspective correct scene outside of a fairly narrow frustum. Maybe they use generative AI to fill the gaps? But what has come out about the format doesn't seem to indicate that, it's still fundamentally a stereo pair just with additional depth data from the depth sensor. I still see gaussian splats as the technique that will finally deliver on practical true spatial photography.