GenAdI: Towards automatic indexing of Russian-Polish church records

Dataset comprised of church book records aimed at End-To-End Information Extraction using Connected-Offline Handwriting Recognition.

This dataset is difficult: given just an image with (usually) several (birth, marriage, death) records on it, recognize the Russian-language text (some items also have Polish language equivalents) and return appropriate Polish language data points (or ISO 8601 dates and periods). This requires not only understanding of underlying Russian text, but also proper normalization and Natural Language Generation.

Given the fact that there are up to 6 records per image, it is important to return data regarding specific record.