My picasa2xmp for immich is a forked version of ckeller42/picasa2xmp, with fixes for python3 and some additions to the xmp output. This is needed, so that immich recognizes the faces originating from the .picasa.ini files. At least in my tests, the .xmp files generated by the ckeller42's version where missing the "AppliedDimensions" section, which immich apparently needs to correctly find faces via the .xmp files.
To add the image dimensions I used the already required exiftool, so no new dependencies. Other dependencies are commented out, as they're apparently not used and I did not find another easy way, of making it work.
Find below the original readme.
picasa2xmp is a command line tool to extract face tag information from picasa ini files and contacts and storing it as xmp sidecar files.
picasa3meta is needed for reading the picasa database. (btw the metaSave tool was used as a template for this script) Additionally exiv2 and exiftool is needed.
The xmp specification describes sidecar files only for raw data. For jpeg images the xmp information is stored in the image. Thats why lightroom does not parse sidecar files for jpegs. This can be a real pain if you have an incremental backup solution. Every metadata modification changes the file and causes an increase in my backup space. Fortunately digikam and darktable also work with xmp sidecar files for jpegs. If you have a image file (e.g. test.jpg) the corresponding sidecar file should be names test.jpg.xmp
It processes all picasa.ini files in the specified folder and stores the updated xmp files in the specified output meta data folder.
> ./picasa2xmp.py --photos <imagefolder> --dest <outmetafolder>
If there already is a xmp file with the original image this file is used and update.
If there is no xmp file a sidecar file from the image is created.
Face rectangles are then added as xmp mwg-rs.Regions.
Additionally HierarchicalSubject tags of the form people|Name are added to the xmp file.
This script is still under testing and is a quick hack. Always make a backup of your data. It currently uses exiv2 and exiftool for adding the xmp data to the files. Would be great to change this to pyexiv2 but I was to lazy to do that yet.