CheckBoxes not recognized

Question

CheckBoxes not recognized

Jack Halpern 20

Using Doc Studio I am able to train for AM or PM checkboxes. See first screenshot. Yet when I run test (or use the API) the field does not appear. Second screen shot. I've even tried testing with the same doc's I used for training. Please advise.

Thanks

Jack

User's image

Mahesh Kurva 9,575 Reputation points Microsoft External Staff Moderator

2025-10-20T14:27:56.6+00:00
Hi Jack Halpern •

It sounds like you're running into an issue where the checkboxes you trained in Doc Studio aren't appearing during testing or in the API response.

Here are some steps you can take to troubleshoot this issue:

Double-Check Training Data: Make sure that the training data for the AM and PM checkboxes is correctly labeled. If there are any inconsistencies in how the checkboxes were trained, it might lead to recognition issues.

Review Input Formats: Verify that the document you are testing with matches the format and requirements outlined in the Azure Document Intelligence documentation. Ensure your document is either a PDF or supported image format (JPEG, PNG, etc.), is within the size limits, and is not password-protected.

Limitations of the Model: Since there's a known limitation with extracting highlighted text, ensure that the fields you're expecting (the checkboxes) are marked in a way the model recognizes. If possible, modify your document to replace highlighted text with checkbox inputs if highlighted text extraction is being used.

Inspect API Responses: Check the API response thoroughly to see if there are any indication of the checkboxes being detected but not returned correctly. It could provide more insight into what might be going wrong.

Testing on Multiple Documents: Since you mentioned you retried with the same documents, try a few different documents with varied structures to see if the model has difficulty with specific layouts.

Model Updates and Support: Ensure that you are using the latest version of the model and API you are working with. Sometimes these features can improve or have fixes that could assist in resolving the issue.

I hope this information helps. Please do let us know if you have any further queries.
SRILAKSHMI C 8,275 Reputation points Microsoft External Staff Moderator

2025-10-22T04:30:59.1+00:00

Hi Jack Halpern,

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!
Jack Halpern 20 Reputation points

2025-10-22T20:49:19.75+00:00

Thanks to both of you for the advice. I reviewed all the suggestions without success. HOWEVER, I deleted my AM and PM fields, re-added the fields, retrained the document. Success! Not sure how I screwed up the initial training but all's well that ends well.

Thanks

Jack
SRILAKSHMI C 8,275 Reputation points Microsoft External Staff Moderator

2025-10-23T04:31:05.5033333+00:00

Hi Jack Halpern,

That’s great to hear I’m really glad that re-adding the AM and PM fields and retraining resolved the issue!

It’s possible that the initial field mappings or labels didn’t fully sync during the first training session, which can occasionally cause extraction inconsistencies. Retraining after re-adding the fields essentially refreshed the model references, so it’s great that it worked perfectly this time.

Thank you for sharing the update I appreciate you taking the time to confirm the resolution!

Since I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted? This helps others in the community with the same question find the solution more easily.

Thank you!

1 answer

Your answer

Mahesh Kurva 9,575 Reputation points Microsoft External Staff Moderator

2025-10-20T14:27:56.6+00:00

Hi Jack Halpern •

It sounds like you're running into an issue where the checkboxes you trained in Doc Studio aren't appearing during testing or in the API response.

Here are some steps you can take to troubleshoot this issue:

Double-Check Training Data: Make sure that the training data for the AM and PM checkboxes is correctly labeled. If there are any inconsistencies in how the checkboxes were trained, it might lead to recognition issues.

Review Input Formats: Verify that the document you are testing with matches the format and requirements outlined in the Azure Document Intelligence documentation. Ensure your document is either a PDF or supported image format (JPEG, PNG, etc.), is within the size limits, and is not password-protected.

Limitations of the Model: Since there's a known limitation with extracting highlighted text, ensure that the fields you're expecting (the checkboxes) are marked in a way the model recognizes. If possible, modify your document to replace highlighted text with checkbox inputs if highlighted text extraction is being used.

Inspect API Responses: Check the API response thoroughly to see if there are any indication of the checkboxes being detected but not returned correctly. It could provide more insight into what might be going wrong.

Testing on Multiple Documents: Since you mentioned you retried with the same documents, try a few different documents with varied structures to see if the model has difficulty with specific layouts.

Model Updates and Support: Ensure that you are using the latest version of the model and API you are working with. Sometimes these features can improve or have fixes that could assist in resolving the issue.

I hope this information helps. Please do let us know if you have any further queries.
SRILAKSHMI C 8,275 Reputation points Microsoft External Staff Moderator

2025-10-22T04:30:59.1+00:00

Hi Jack Halpern,

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!
Jack Halpern 20 Reputation points

2025-10-22T20:49:19.75+00:00

Thanks to both of you for the advice. I reviewed all the suggestions without success. HOWEVER, I deleted my AM and PM fields, re-added the fields, retrained the document. Success! Not sure how I screwed up the initial training but all's well that ends well.

Thanks

Jack
SRILAKSHMI C 8,275 Reputation points Microsoft External Staff Moderator

2025-10-23T04:31:05.5033333+00:00

Hi Jack Halpern,

That’s great to hear I’m really glad that re-adding the AM and PM fields and retraining resolved the issue!

It’s possible that the initial field mappings or labels didn’t fully sync during the first training session, which can occasionally cause extraction inconsistencies. Retraining after re-adding the fields essentially refreshed the model references, so it’s great that it worked perfectly this time.

Thank you for sharing the update I appreciate you taking the time to confirm the resolution!

Since I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted? This helps others in the community with the same question find the solution more easily.

Thank you!

Answer 1

Hello Jack Halpern,

Welcome to Microsoft Q&A and Thank you for reaching out.

In addition to the points shared by Mahesh Kurva, here are some additional insights and best practices to help address this issue.

If your checkboxes are faint, scanned lightly, or not uniformly printed, the model may fail to detect them. Increasing contrast or using digital versions of the forms can improve detection accuracy.

When labeling in Document Studio, ensure your checkbox labels don’t overlap nearby text or borders. Overlapping regions can cause the model to misinterpret the selection mark area.

If documents are slightly rotated or skewed during scanning, selection mark detection can fail. You can preprocess the document using OCR tools or the Layout API to normalize orientation before analysis.

If your checkbox fields were renamed or modified after training, the API response schema might not reflect the changes. Make sure the field names in your inference request match the schema defined during training.

Use the REST API endpoint (instead of Studio test) with "features": ["selectionMarks"] explicitly included in your request body. This helps confirm whether the marks are detected at the raw extraction level.

If you’ve trained multiple model versions, test using the previous working version to see if the issue appeared after a specific update this helps isolate model drift or labeling regressions.

If your checkbox area also includes text (e.g., “☑ AM”), ensure you didn’t label it as both a text field and selection mark. Azure Document Intelligence might prioritize text extraction and skip the mark.

Please refer this Troubleshoot Highlight Text Extraction in Azure Document Intelligence

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Share via

CheckBoxes not recognized

1 answer

Your answer