Form Recognizer batch processing and model enrichment ?

Question

Form Recognizer batch processing and model enrichment ?

Développeur Le Studio 31

Hello, I had some questions about form recognizer and the pre-built model of invoice analysis:

If we submit a batch of invoices, is the software able to determine at which page an invoice starts and stops?
Is it possible to correct the results of the analysis or to add them if they are not relevant (enrichment of the model)
On the custom model, is there a way to train the model directly through an API (without form recognizer studio) ?
Thanks for the answer

Answer 1

romungi-MSFT 29,711 Microsoft Employee

@Développeur Le Studio You can pass a document with upto 2000 pages to the prebuild model or the API where you can have multiple invoices and the API should provide the result of all invoices provided they are of the supporting format and the response should be available in the documentResults section of the response with different page numbers. A quick way to test this scenario is to use the API from the studio or the REST API.

It is currently not possible to tweak a pre-built model or enrich a model. You will need to create a custom model and then create different versions of your custom model and compose them under a single model id to analyze different documents.

Yes, you can train the model using the Build Model API for FR v3.0 onwards.

If an answer is helpful, please click on or upvote which might help other community members reading this thread.

Développeur Le Studio 31 Reputation points

2022-06-22T07:35:39.567+00:00

Thank you for your responsive response. I have a doubt about the batch part. We were able to test with the invoice prebuild model batches of several invoices and it recognizes almost nothing, only 1/2 pages and still the results are not conclusive. For example, on this screeshot the analysis stops at page 3
romungi-MSFT 29,711 Reputation points Microsoft Employee

2022-06-22T08:08:01.613+00:00

Are you using free tier for your resource? A free tier resource request can only analyze the first 2 pages of a document. This is documented here in the documentation.

For PDF and TIFF, up to 2000 pages can be processed (with a free tier subscription, only the first two pages are processed).
Développeur Le Studio 31 Reputation points

2022-06-22T08:31:58.727+00:00

Ok, thanks a lot !
Dyt13 201 Reputation points

2022-12-31T16:51:40.43+00:00

Hi @romungi-MSFT , I've tried for the last couple of days to analyze multiple documents in one post request without success unfortunately. Can you please elaborate on how you managed to obtain

the result of all invoices [...] in the documentResults section of the response with different page numbers

I'm using Form Recognizer REST API (https://westus.dev.cognitive.microsoft.com/docs/services/form-recognizer-api-2022-08-31/operations/AnalyzeDocument)

Basically when I upload a 2 pages PDF (composed of two distinct 1 page document to be analyzed) I expect the analyzeResult.documents (typecast = list in python) to be a list with two elements. It's not the case.

That being said I indeed observed that for some elements such as analyzeResult.pages and analyzeResult.tables the Form Recognizer indeed works great.
Unfortunately what I'm interested in is gathering the information which lies in the fields attributes of each element available in analyzeResult.documents.

I've opened an issue on GitHub (https://github.com/Azure/azure-sdk-for-python/issues/28126)

If the feature is not available yet do you think that this feature is on the roadmap for further releases ?

Thanks in advance for your help
romungi-MSFT 29,711 Reputation points Microsoft Employee

2023-01-02T12:54:25.36+00:00

@Dyt13 I believe the document you uploaded contains two pages where you are expecting the response to have each page result as a single document response.
I think this is not possible with the current API as the FR API excepts a single document for each request and provides result of each page as page numbers. The guidance for such scenarios has been to split the document and pass each page in individual requests. Please see this thread from SO for details.
Dyt13 201 Reputation points

2023-01-02T17:38:32.333+00:00

Got it thanks for the confirmation. Would be a really nice feature to have -> batch processing :)

Form Recognizer batch processing and model enrichment ?

0 additional answers

Activity