Unable to indicate checkbox

Question

Hello, I came across the same issue on Stack Overflow related to Azure AI services Document Intelligence. I am experiencing a problem while training a model to read a form. The model is unable to indicate if a box is selected or not. I was wondering if I can manually modify the *.labels.json and *.ocr.json files to indicate that the box is selected and thereby improve the accuracy of the model. Alternatively, if you have any other suggestions, please let me know. I have attached a sample image that the model is unable to interpret. Thank you. Sample-type.PNG

Answer 1

@Hermoso, Eric I think the recommendation to update the backend files was from a previous version of the service i.e Form recognizer, updating the files at backend might fail the model training with the current version. I would recommend following the guidelines for labeling checkboxes as quoted from this document.

Labeling values is required. Don't include the surrounding text. For example when labeling a checkbox, name the field to indicate the check box selection for example selectionYes and selectionNo rather than labeling the yes or no text in the document.

I understand in your case the checkboxes were not automatically detected as indicated in the below screen shot if they were detected.

If that is not the case, you can introduce a label to highlight the checkbox and provide a name like "LeftAsFoundYes" "EstablishedNo" "RebuiltNo" and then use the "Draw Region" button to start highlighting the box next to the text and label them with the labels created.

For example, I have used this box and region next to it to label the boxes in my sample document.

Mapped them to the following labels after the region is drawn.

This can train the model accordingly and you can avoid manually updating the files at the backend. Thanks!!

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.