It can be seen from the video (from the way they avoid it) that the "augmentation" is always superimposed on the "reality". I.e. someone can't walk in front of the virtual objects you put on the table.
> It can be seen from the video (from the way they avoid it) that the "augmentation" is always superimposed on the "reality".
I think you mean “inferred” rather than “seen”, if it is an assumption based on avoidance, and there are other explanations; while HoloLens is better equipped than phone-holder software AR to avoid this, the one time I did get to use one there were some glitches when the “augmentation” should be obscured y the “reality”. If ARCore handles that, in principle, but is currently annoyingly glitchy in practice on its current preview-quality state, you might reasonably avoid it in demos.
You don't need to reliably get a good take to produce a video; you need one good take.
If it's bad enough that it won't look right even after taking the best of a large number of attempts, that's as good as the feature not existing for the purpose of my question.
Is that a limitation of ARKit too?
What would it take to make it "real 3D"?