In Azure Form Recognizer service, can we import custom trained model into pre-built model?

Question

We are evaluating Form recognizer to be used for our receipts data capture use case for our platform in which receipts from 1000's of US based seller's are captured. We are finding that we will have to use custom trained models in many cases.

The question we have is: after training a custom model for a particular receipt type, can this custom model be incorporated/imported into Pre-built receipt model and used as pre-built model? The reason this question is raising is custom model is expensive to use compared to pre-built models.

Thank you in advance for your time and effort,

Anup

Answer 1

No, it is not possible to import a custom trained model into a pre-built model in Azure Form Recognizer. Pre-built models and custom trained models are separate entities within the service and cannot be combined or merged.

Pre-built models in Form Recognizer are trained by Microsoft and are designed to extract information from common document types like receipts, invoices, and business cards. These models are optimized for specific document types and are trained on a large dataset to provide accurate extraction results.

On the other hand, custom trained models are trained using your own labeled data and are tailored to extract information from documents with specific layouts or structures that are unique to your business needs.

While the cost of using custom trained models may be higher compared to pre-built models, it offers the advantage of greater flexibility and accuracy for your specific document types.

If you require the capabilities of a pre-built receipt model but also need to capture information from other types of receipts, you can consider using a combination of pre-built models and custom trained models. You can use the pre-built receipt model for common receipts and use custom trained models for specific receipts that require additional processing.

It's important to evaluate your specific requirements and consider the trade-offs between using pre-built models and custom trained models to determine the best approach for your use case.