Dear Forum,
I've been steadily reading myself through the documentation for the form recognizer for the past weeks and I think I'm ready to ask questions.
I am currently working with a type of document that consists of a document header, a main part containing a table and a footer. In the table there are 2 possible variations which shift the layout a little. As far as I understand it, this means I am having a usecase for a composed custom template model. One model each for the different table variations with about 5-15 (minimum) training docs each, composed together. Did I get that right so far?
My real question would be: How to go about training those 2 models within one project? As far as I know I can only provide a connection to one storage container, so I can only have the training data for one of the table variations in my blob storage and need to clean it out whenever I switch between variation submodels for retraining. This sounds quite tedious and I was wondering if there are other options like keeping the variation submodels in different projects for training and exporting them into a third one where they will be composed and used?
Is it even possible to use models saved in different projects using the form recognizer studio?
Or am I thinking completely wrong? How would one go about handling a composed model whose sub-models require semi regular retraining?
Apologies in advance for any ignorance on my part as I've just started learning to work with azure.
Thanks a lot for any and all help!
or upvote
which might help other community members reading this thread.