Document intelligence layout-model doesn't detect tables for .xlsx

[hamer101]

• Package Name: azure-ai-documentintelligence
• Package Version: 1.0.0b1

No tables are detected for a .xlsx file - the poller's result field of tables remains empty.

Although I suspected loosely written tables might not get detected, I assumed that areas that have been declared as "tables" in the excel application will get extracted.

The analyzed .xlsx file contained one such defined table, along with few subtotal fields. Code used:

with open('/tmp/example.xlsx','rb') as f:
    poller = document_intelligence_client.begin_analyze_document(
        model_id="prebuilt-layout",
        analyze_request=f, 
        content_type="application/octet-stream"
    )
    form_recognizer_results_markdown = poller.result()
print(form_recognizer_results_markdown.tables)

I was about to consider it a feature request, but given it some thought, it deviates quite a bit from behavior I expected :)

[YalinLi0312]

@hamer101 Thanks for your contact, we'll investigate asap!

[YalinLi0312]

Hi @hamer101 , unfortunately we are not supporting the feature so far, it is saying "Table is not supported if the input file is XLSX." in our doc(link). I'll tag the service team to aware your request, thanks!

[github-actions[bot]]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @ctstone @vkurpad.

[maltenlz]

Is there any update on this? This feature be very appreciated!

[YalinLi0312]

Add @bojunehsu to bring more attention on this feature request on layout-model.

[bojunehsu]

@hamer101 Thanks for the feedback. Can you please share more information about the intended use scenarios to help us prioritize the work? Thanks.

[hamer101]

I wanted to feed a xlsx to a LLM via a pipeline that already deals with tables by tables parameter of layout-model's result.

[bojunehsu]

@hamer101 Thanks for sharing the scenario. We will add this request to our backlog.

[YalinLi0312]

@hamer101 Since this is not an SDK issue, will close it. Thanks

[clausagerskov]

@bojunehsu Where can we track this feature request?