AI Model should detect different form types and orientation of PDF document

[sgggt]

For example, if my industry has two different types of standard invoice forms that almost everyone uses, the AI invoice processing model should be able to detect which form was uploaded and use either AI predict model 1 for standard form type 1 or use AI predict model 2 for standard form type 2.

Until the AI models can differentiate between different form types uploaded, maybe one way would be to detect the orientation of the form? Landscape forms could be form type 1, horizontal forms could be form type 2.

[JoeF-MSFT]

Thanks for posting this idea @sgggt. This is planned and targeted to be released in a few weeks from now. 🙂

[sgggt]

Excellent @JoeF-MSFT can you provide a little more context what's being implemented? Thank you!

[JoeF-MSFT]

Sure, it will work like this:

1. You start by creating a Form Processing model as you do today.
2. On the first step of the training wizard, you will be asked to specify the fields you want your model to extract.
3. Then you upload at least 5 sample documents for each document format you want the model to learn how to extract information.
4. You tag the fields you have specified across all the documents you have uploaded to teach the model.
5. The model gets trained.
6. Once the model is trained, you can use it in a flow or in an app as you do today. The model will automatically know where to look for the fields for the different document formats you have used for training.

Hope this gives a preview of what to expect. Let me know if you need any additional info. 

[sgggt]

@JoeF-MSFT sorry for my ignorance, but isn't that how it's done currently? 

[JoeF-MSFT]

Hi @sgggt,

Thanks for the question.

Today if, for example, you want to automatically process invoices from three separate vendor where every vendor has a different document layout/format you will need to build three different AI Builder form processing models, one for each vendor.

With this upcoming update, you will only need to build one form processing model and you can use samples with different layouts to train a single model.

Hope this clarifies. 🙂 If not, happy to answer any additional question. 

[JoeF-MSFT]

Good news! Now you can train a single form processing model to detect different document types: https://flow.microsoft.com/en-us/blog/ai-builder-form-processing-models-can-now-train-using-document...