how do I add multiple drawing box to label missed item number on a drawing using custom extraction model

Kimberley Attwell 0 Reputation points
2025-08-03T23:43:01.4966667+00:00

I am trying to extract Item numbers on an exploded drawing. Most of the numbers are recognised so I can assign them to a field. But there are some item numbers that are not recognised. it I draw a bounding box around one of the missed numbers it only allows one per field. and I can't add another field with the same name.

how do I get the custom models to recognise all numbers and ensure it does not miss some of them?

Azure AI Document Intelligence
0 comments No comments
{count} votes

1 answer

Sort by: Most helpful
  1. Alex Burlachenko 13,330 Reputation points Volunteer Moderator
    2025-08-04T08:50:48.25+00:00

    hi kimberley! thanks for posting this ))

    u can actually train the custom extraction model to catch those missed numbers. the key is to use 'arrays' in your schema.

    go to your custom model in the azure portal. when defining the field for item numbers, make sure to set it as an 'array' type instead of 'string'. this tells the model 'hey, expect multiple values here!'

    then, during training, draw bounding boxes around ALL the item numbers u want it to recognize - even the ones it missed. the model learns better when u show it both correct and missed examples. https://docs.microsoft.com/en-us/azure/applied-ai-services/form-recognizer/label-tool

    sometimes the model misses numbers because of image quality. try enhancing the contrast or resolution before uploading. this might help in other tools too )

    worth looking into the 'cross page validation' feature if your drawing spans multiple pages. the model can sometimes catch missed items by comparing similar patterns across pages.

    when labeling, vary your bounding box sizes slightly for the same item number. this makes the model more flexible in real world scenarios.

    if u keep having issues, try reducing the 'confidence threshold' temporarily during testing. this forces the model to show u more potential matches, even if it's not super confident. u can filter false positives later )

    bonus thought - sometimes the issue isn't the model but the schema design. make sure u didn't accidentally mark the field as 'single value only' during setup. its an easy miss!

    let me know if u hit any snags

    Alex

    and "yes" if you would follow me at Q&A - personaly thx.
    P.S. If my answer help to you, please Accept my answer
    

    https://ctrlaltdel.blog/

    0 comments No comments

Your answer

Answers can be marked as Accepted Answers by the question author, which helps users to know the answer solved the author's problem.