You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While trying to create markdown or text files from AWS Textract JSON output using the get_text_from_layout_json function, the contents of ALL the list items are duplicated in the output.
Expected Behavior
Each list item's contents should be included in the output only once.
Layout elements can also point to different objects, such as TABLE objects, Key-Value pairs, or LAYOUT_TEXT elements in the case of LAYOUT_LIST
Due to this, when getting all layouts from the Textract JSON output (LinearizeLayout._get_layout_blocks), the LIST_LAYOUT as well as its child TEXT_LAYOUT layout elements are included, which leads to the duplication in output text.
The get_text_from_layout_json function is a wrapper over LinearizeLayout.get_text function which loops over all layouts (blocks with LAYOUT.* type) from the Textract JSON output and collects the text contents from their children blocks.
The fix lies in the LinearizeLayout._get_layout_blocks function where we can exclude the LAYOUT_TEXT elements which are children of LAYOUT_LIST elements.
I've made the proposed changes here. I'd be happy to create a PR!
Also updated the min repro repo to use this latest version and the comparison between the fixed branch and main shows the change in the sample output, i.e. list content is not duplicated anymore.
Found a couple other cases where LAYOUT_LIST had LAYOUT_SECTION_HEADER and LAYOUT_TITLE as its children. Created a PR with a fix that excludes all LAYOUT* elements which are children of LAYOUT_LIST elements.
Sure! I can't share the original document, but here's another sample document which recreates the LAYOUT_LIST -> LAYOUT_SECTION_HEADER scenario. JSON output for reference.
You can check this comparison to see the difference between ignoring text only v/s all layout children.
Current Behavior
While trying to create markdown or text files from AWS Textract JSON output using the
get_text_from_layout_json
function, the contents of ALL the list items are duplicated in the output.Expected Behavior
Each list item's contents should be included in the output only once.
Related Issues
Possible Solution
The AWS docs on Textract Layout Response Objects mention that in the case of
LAYOUT_LIST
elements, their children can point toLAYOUT_TEXT
elements, which is the case here.Due to this, when getting all layouts from the Textract JSON output (LinearizeLayout._get_layout_blocks), the
LIST_LAYOUT
as well as its childTEXT_LAYOUT
layout elements are included, which leads to the duplication in output text.The get_text_from_layout_json function is a wrapper over LinearizeLayout.get_text function which loops over all layouts (blocks with
LAYOUT.*
type) from the Textract JSON output and collects the text contents from their children blocks.The fix lies in the
LinearizeLayout._get_layout_blocks
function where we can exclude theLAYOUT_TEXT
elements which are children ofLAYOUT_LIST
elements.Steps to Reproduce
Minimal reproduction repo: adityachandak287/textractprettyprinter-list-duplication-bug-repro
The repository contains the following for reference:
Environment
Edit: Added related issues section.
The text was updated successfully, but these errors were encountered: