Textract: pretrained tables?

Question

Hello,

in the pricing page for textract (https://aws.amazon.com/textract/pricing/) I see this mention of "pretrained tables". What are they? I mean, other than the obvious. I can't find any reference or documentation on how to make use of these. So far, I've been getting my raw tables from the analyze document response and then working my way through them and their data programmatically. I know we can and indeed we do use custom queries, but queries are not good for tables, so I wonder is these pretrained tables are a feature that could make my work less dependent on parsing & reading really specific code for tables.

Thanks!

Giovanni Lauria · Accepted Answer

"Pretrained tables" in Amazon Textract refers to its ability to automatically extract tabular data from documents. This feature is part of the AnalyzeDocument API and can identify various components of a table. Enhancements introduced in April 2023 allow Textract to automatically detect titles, footers, section titles, and summary rows in documents. The extracted table data is returned as Block objects in the responses. However, Textract does not interpret the data or understand the context of the table, which would need to be implemented in your application.

Textract: pretrained tables?

Relevant content