Open hybridherbst opened 1 year ago
Maybe we can expose something like the body API which returns the positions of head, torso, arms, etc?
In practical terms APIs (ARKit / ARCore) nowadays usually let the developer choose between a "fully-body mode" (no face blendshapes) and "face mode" (limited body, usually only head). Not sure how feasible mixing everything into one proposal would be.
I think they would have to be separate specs. Body would contain the person's physical characteristics so we have to be very careful when exposing those.
For Quest Pro and similar "inside facing" devices I see how only the facial blendshapes are needed, because head pose equals device pose.
However, for phone-based usecases such as face tracking on Android or iOS, an additional head pose (pos, rot, scale) is needed to match from screen-based session to a reasonable head model.