In this article, you will gain insights into the following:

• Understanding interactive bounding boxes
• Changes implemented on the validation screen

What are interactive bounding boxes

During the process of validating data extracted from documents, you may have noticed that the values on the document are surrounded by blue borders. These borders, also known as bounding boxes, serve the purpose of selecting values on the document. To enhance the annotation process and make it more intuitive, we are introducing a new and improved version of bounding boxes.

Changes on the validation screen

Interactive bounding boxes

• Hovering over an bounding box enables you to check which field holds the corresponding value.

• Clicking on an extracted value redirects you to the associated field.

• When a field is selected, clicking on another extracted value on the document redirects you to the field linked to that value, rather than changing the value in the selected field.

• Adjusting the size of a bounding box is now easier, as you can modify it using the corners.

These enhancements significantly simplify and expedite the annotation process.

Highlighted values with different colors

• Green: Values predicted with high confidence, surpassing the threshold set for the field, and values manually validated by annotators.

• Yellow: Values predicted with confidence below the threshold set for the field.

• Red: Values that prevent document confirmation, such as incorrectly formatted extracted data.

This change provides valuable insights into the predicted data. You can promptly identify values that hinder document processing or fields that require additional annotations to improve AI confidence.

Suggested bounding boxes

If you wish to review other recognised values on the document available for annotation, you can easily display suggested bounding boxes. Simply hold down the SHIFT key on your keyboard.

By incorporating these improvements, the annotation process becomes more user-friendly and efficient than ever before.

This feature is currently at an early stage and is only available to clients who have registered to participate in the test phase. If you have any questions please reach out to us – support@rossum.ai.