Picasa, Digikam, Immich, Tonfotos, etc. all use XMP to write the face data to XMP sidecars. Being able to read the sidecar file for faces, and the images themselves for embedded data, would save a lot of time for images that used the more standardized approaches to face annotation in the past.