Hello Etienne,
PrincipalPointU and PrincipalPointV in the exported XMP file are the relative offsets from the center of an image in x axis and y axis respectively. Principal point x and Principal point y in the application are offsets from the center of an image in [mm] w.r.t. 35 mm film format. Simply said, the values in the XMP file are the values in the application divided by 35 mm. Therefore, these values are dimensionless.
You are right about the description in the application, we will change that to make it clear.