Azure Form Recognizer
An Azure service that applies machine learning to extract text, key/value pairs, tables, and structures from documents.
696 questions
Form Recognizer Studio 3 - Cannot use custom model to help in labelling new documents
D. Lotti - Gruppo Finservice
6
Reputation points
Hello,
in the previous tool (FOTT preview) there was the GREAT feature to use a trained model to label data for new documents.
This function helped me a lot and saved me many hours.
I can't find it in the new form recognizer studio 3 and I miss it very much!
I think a lot of user consider that function very valuable.
I would like an option to use a model when I upload some documents in the labelling section.
I expect to find a lot of fields labeled and I verify and correct wrong ones.
Now I have to label each document from zero and it's quite stupid.
Thank you
1 vote
-
YutongTie-MSFT 27,761 Reputation points
2022-08-31T22:18:51.223+00:00 Hello
I think you are mentioning the custom model feature? It is in Studio v3 as below screenshot - https://learn.microsoft.com/en-us/azure/applied-ai-services/form-recognizer/concept-custom
Is this what you want? If not, could you pleasse share a screenshot of FOTT?
Regards,
Yutong
Hi @YutongTie-MSFT ,
I think @D. Lotti - Gruppo Finservice is talking about the 'Auto-label' feature that you can see under the 'Actions' menu in the FOTT's Tags Editor
Which I agree, it is quite useful, and it saves us a lot of time.
Regards,
Houcem.
Yes exactly.
Are you able to still use the old tool to label and then switch to the new one for training with the new version of the models?
Yes, I'm using Form Recognizer Studio v3 with custom model.
This is the evolution of the previous tool that was called "Form Ocr Testing Tool (FOTT)"
https://fott-2-1.azurewebsites.net/
The old tool prompts to switch to the new one but the new one lack the feature to use the custom model to label new documents.
Will it be implemented in the near future?
No, the Form Recognizer Studio (v3) uses a fields.json for the tags, whereas FOTT (v2.1) uses a {ProjectName}.fott file. Two different json structures.
Hello @D. Lotti - Gruppo Finservice @HoucemMEZHOUDI-8102
I got you both, let me check with product manager about this feature.
Regards,
Yutong
Any news from the product manager?
The feature will be included in a next build or not?
Thank you
Hello @D. Lotti - Gruppo Finservice
There is no update since the long weekend in US, I will follow up with they again today, Thanks for your patient.
Regards,
Yutong
In the old version there is also the option to "Edit and Upload to Training Set" in the Analyze tab. This brings in all the values it was able to tag automatically and saves time vs doing it from scratch. This is not found in the new version.
Any news from the product manager?
Just to know if I have to wait days, weeks or months or if you force me to switch to another ML tool :-(
Thank you
Two months have passed and no reply.
Can you please check with the product manager? I would like at least to know if the feature is on the roadmap or not!
Am I wainting for nothing?
Thank you
Hello @D. Lotti - Gruppo Finservice
Sorry I missed your message. I think they are still considering this feedback, but I just popped it up and hope they can give us a yes or no answer today or tomorrow since we have provide a long time for them. Will let you know.
Apologize for the late reply, please mention me for any more questions.
Regards,
Yutong
Hello @D. Lotti - Gruppo Finservice
I just got the news from the pm, the auto-label feature will be available in the Jan/Feb 2023. Sorry for the delay.
I hope this helps.
Regards,
Yutong
Hello @YutongTie-MSFT,
Any news on when will this feature be available?
Thanks a lot for your help!
Regards.
Sign in to comment
1 answer
Sort by: Most helpful
Hi yutong
I love this feature too. It helps a lot for quick labeling. I would upvote this idea. New version should include all good features.
Sign in to comment
Activity