When reviewing auto-labeled documents in the HITL or labeling training interface, I'm not clear from Google's instructions whether to identify empty fields on a form. For example, if I have these fields:
Should I leave them unlabeled since they don't have values, or draw the bounding boxes that identify where a value would be and leave the text value blank like this?
I'm reviewing auto-labeled forms and seeing these blank fields unlabeled frequently, and I'm not sure if I'm helping to improve the model by drawing the bounding boxes.
Confirmed with the Product Managers, you should leave empty fields unlabeled when creating training data for a Custom Processor (Uptraining or Custom Document Extractor)