“Custom Extraction Model” it sometimes merges 2 rows data into 1

Question

“Custom Extraction Model” it sometimes merges 2 rows data into 1

Dmytro Halahan 10

Hi

I’ve started to explore Form Recognizer and come across the following issue.

When I train a “Custom Extraction Model” for receipts it sometimes merges 2 rows of data into 1.

However, this problem doesn’t occur when using a prebuilt receipt model. Also, it feels like the table region is recognized differently within the prebuilt model.

The question:

Is it possible to repeat the prebuilt model region recognition in the custom model?

If yes, could you kindly provide some guidance?1.png

The example picture is attached.

Kind regards,

Dmytro Halahan

santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-19T11:58:57.17+00:00

Hi @Dmytro Halahan ,

Thanks for sharing the details on Microsoft Q&A forum!

I understand that you are experiencing issues with the model merging two rows of data into one. Iam happy to assist you in this regard.

I tried to reproduce using custom model from my end with the given document and another document, it is working perfectly. Please look into the below screen shots.

I suggest you upload the clear picture of receipt or document to the Test model.

I hope this helps! Please try and let us know. Thank you.
santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-20T11:34:01.1966667+00:00

Hi @Dmytro Halahan ,

Following up to see if the above suggestion was helpful. And, if you have any further query do let us know. Thank you.
Dmytro Halahan 10 Reputation points

2023-10-20T16:10:21.1066667+00:00

Hi. It was helpful. But could we go a little bit deeper into details. Does AI has the sensitivity to the image properties? if yes what properties count here (angles, light, etc.)?
santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-24T13:41:35.4333333+00:00
Hi @Dmytro Halahan ,

Thanks for the response. The input requirements for Document Intelligence.

For best results, provide one clear photo or high-quality scan per document.

Image dimensions must be between 50 x 50 pixels and 10,000 px x 10,000 pixels.

Please look into the documentation for more details: Build and train a custom model - Document Intelligence (formerly Form Recognizer) - Azure AI services | Microsoft Learn
santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-25T13:25:01.7+00:00

Hi @Dmytro Halahan ,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution, please do share that same with the community as it can be helpful to others. Thank you.

1 answer

Your answer

santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-19T11:58:57.17+00:00

Hi @Dmytro Halahan ,

Thanks for sharing the details on Microsoft Q&A forum!

I understand that you are experiencing issues with the model merging two rows of data into one. Iam happy to assist you in this regard.

I tried to reproduce using custom model from my end with the given document and another document, it is working perfectly. Please look into the below screen shots.

I suggest you upload the clear picture of receipt or document to the Test model.

I hope this helps! Please try and let us know. Thank you.
santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-20T11:34:01.1966667+00:00

Hi @Dmytro Halahan ,

Following up to see if the above suggestion was helpful. And, if you have any further query do let us know. Thank you.
Dmytro Halahan 10 Reputation points

2023-10-20T16:10:21.1066667+00:00

Hi. It was helpful. But could we go a little bit deeper into details. Does AI has the sensitivity to the image properties? if yes what properties count here (angles, light, etc.)?
santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-24T13:41:35.4333333+00:00

Hi @Dmytro Halahan ,

Thanks for the response. The input requirements for Document Intelligence.

For best results, provide one clear photo or high-quality scan per document.

Image dimensions must be between 50 x 50 pixels and 10,000 px x 10,000 pixels.

Please look into the documentation for more details: Build and train a custom model - Document Intelligence (formerly Form Recognizer) - Azure AI services | Microsoft Learn
santoshkc 15,600 Reputation points Microsoft External Staff Moderator

2023-10-25T13:25:01.7+00:00

Hi @Dmytro Halahan ,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution, please do share that same with the community as it can be helpful to others. Thank you.

Answer 1

Dmytro Halahan 10

Sorry, the image attachment looks strange in the main message.

Please find the correct one below.

Results of Custom Model

Results of Prebuilt Model

Share via

“Custom Extraction Model” it sometimes merges 2 rows data into 1

1 answer

Your answer