We have a custom photo booth for taking photos of people for use with photogrammetry - the usual vertical cylinder of cameras with the human subject stood in the middle.
We've found that often the lower legs of the subject are missing - this is particularly likely if the subject is wearing dark pants.
The API for PhotogrammetrySession is really very limited, but we've tried all the combinations or detail and sensitivity and object masking we can think of - nothing results in a reliable scan.
Personally I think this is related to the automatic isolation of the subject, rather than the photogrammetry itself. Often we get just the person, perfectly modelled. Occasionally we get everything the cameras can see - including the booth itself and the room it's in! But sometimes we get this footless result.
Is there anything we can try to improve the situation?