Fine tune prebuilt model

Question

I want to use From Recognizer to read invoices but the prebuilt model doesn't pick up the Tax. Is there any way I can train a copy of the already built model so it can detect these fields without having to train a whole model form the ground-up? The model has to be able to read all the fields of invoices with different formats, and training a model for each format isn't feasible. Thanks, Lluís.

Answer 1

@Lluís Gregori Thanks for the question. The Prebuild Models of Form Recognizer can't be trained. you can train a custom model to detect new fields using Form Recognizer. To create a custom extraction model, you can label a dataset of documents with the values you want extracted and train the model on the labeled dataset. You only need five examples of the same form or document type to get started. This should allow you to detect the Tax field on your invoices without having to train a whole model from the ground-up.