Yes, part of the split/merge, as you noted, is the geographic binning of images to ensure proper overlap between the submodels.
I think if you need the split/merge pipeline, it might be beneficial to do your post-processing to get geolocation information.
If not, you could always manually chunk groups of images into separate tasks and then combine in post, but be aware that a naive reconstruction may not have the level of detail you require as the GSD estimate may not be anything close to what it really was in-situ.
Thanks a lot for the answer, super clear and makes sense.
As our picture are shallow under water (archeology orthophotos), we are devising a solution to post label the photos with GPS coordinates. At least that part looks like working.