Slow performance - size limitation?

Question

Hi, I'm using a custom neural model to extract tables from pdfs that are about 50 pages long. One of the tables is quite large and spans 20+ pages, so the table has well over 1000 rows. The problem is that the labeling process becomes very slow after the table has 500 or so rows. After 1000, it can take a few seconds or even 20 seconds at times to insert, edit, or label a row of data in the table.

Is there a limitation to the number of rows a table can have? Or is there a Best Practice for this type of issue? I have tried editing the pdf to limit the number of pages, but this leads to poor test accuracy because each page can have uniquely different information listed in the table.

I am using the latest API version released on 2024-02-29 and have tried it on the 2023-10-31 preview as well.

Answer 1

Hello @MeghanZiegler-1345 , there is a feature which can help to auto-label the table fields to reduce the manual effort.  Can you try this feature and see if this can help the labeling efficiency?  

1. For each page, add/create the "Table" field first as usual to define the field info.
2. After "Run layout" is done, a "table" mark will show up. Click on the icon/mark.
3. A table with the recognized content will show up.  Scroll to the bottom and click "Auto label" and choose the table field accordingly.
4. A "Table labeling" wizard will show up and you can follow the steps to edit and review the rows/columns accordingly.

Here are the Label tips and tricks for Document Intelligence Studio: https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-custom-label-tips?view=doc-intel-4.0.0#auto-label-tables