Can OCR labels be fixed without retraining?

Good morning. I have trained an OCR model and mistakenly used the base model “Utility Bill” when I should have used “Invoice.” I changed it, but the “Document Type” still appears as “Utility_bills.” I would like to know how to correct this without retraining everything. Is it possible?

Second question: My invoices are from Argentina, a type of invoice not listed by UiPath. What is the impact? Can a specific model be trained for Argentina?

@Matias_Clemente.Arg

Not sure about the changing document type name but are you using modern project or classic DU?

The fields won’t be as per country specific and also the extraction will be as per the training data of the model which might miss some Argentina specific fields of any.

Use the Invoices base model without any country specific code.

i don’t know. I always strive to utilize the modern approach, yet I cannot recall the system ever inquiring about it at any point.

@Matias_Clemente.Arg - does the model perform well enough even if you started from the utility bills? do you have all the fields and data you need?

If you worry about the name, you can always use IntelligentOCR - case in which you control not only the name of the doc type, but which fields and with what names and what display settings to be used (throughout from classification / classification validation / extraction / extraction validation).

We are also working on bringing some similar functionality to the cross-platform activity package (ExtractDocumentData activity), but it is a WIP.

Hello, I conducted some tests and was genuinely surprised because there were fields that were quite clear, yet they were not recognized correctly.

@Matias_Clemente.Arg - the utility bills model has quite different fields and training data than invoices… could you try to create a new project with the proper base model?

Hello dear, well, of course I can do it. However, my question aimed at whether there was a way to modify the base model without retraining everything, and if there was any difference in using a base model not specifically designed for the indicated purpose. Your suggestion provided answers to both.
Thank you!.


Which is the superior option for OCR?