Most research papers reporting results on IAM database use different splits of the data into train/test/val than those provided with the database. These are those data splits shared by Théodore Bluche from Human Language Technology and Pattern Recognition lab, RWTH Aachen University. It will create 6482 train line images, 2915 test line images and 976 validation line images.

It contains the page XML name. For example:

a01-000u
a01-000x
a01-003