Authors: Thomas CONSTUM, Lucas PREEL, Théo LARCHER, Pierrick TRANOUEZ, Thierry PAQUET, Sandra BREE

Affiliation: LITIS

Description: We adapt the DAN for joint Handwriting Text Recognition and Named Entity Recognition on full pages.
To do so, we encode the named entities in the ground truth by using special tokens placed after each named entity to extract. We use single separate tags meaning that for each named entity type, there is a tag encoding the person and a tag encoding the category.
Regarding the training procedure we first train the DAN for HTR before training it for HTR+NER.