While traditional OCR products will often include a bounding box reference, Extend uses a mix of multimodal large language models and traditional vision models in order to build our AI powered document processing, and because of this mixture of models including bounding box references is not always possible and coverage of fields is not guaranteed, however we are always working to improve the coverage.

The default bounding box using heuristic based matches only supports the following field types:

  • date fields
  • string fields
  • signature fields
  • array fields (on nested string fields)
  • object fields (on nested string fields)

If you have selected “Advanced bounding box” in the extraction settings in the Extend Studio, you will be able to use bounding boxes for all additional field types. These bounding boxes will have significantly higher coverage.

  • enum fields
  • number fields
  • boolean fields
  • null fields - If a field is declaratively null (e.g. an empty form input) this will be returned as a bounding box reference. If there is no declarative indication of null, bounding boxes will not be returned.

You can toggle this on in the Advanced Settings of an extraction configuration in Extend Studio:

Right now bounding box references are only available for Extract output fields, and are only supported for the following file/document types:

  • PDF
  • IMG (jpeg, png, etc)

Demo

When reviewing, as you use your arrow keys to navigate through each review field, our file viewer will automatically scroll to the relevant page, and if a bounding box reference was made available in our system, it will be highlighted for you to serve as visual aid when review.