Using the bounding box tool

Bounding box

A bounding box is a rectangle positioned on the object.

To annotate using a bounding box, select an object class and then do one of the following:

Click and drag the cursor on the asset and then release the cursor.
Select annotation vertices with two mouse clicks. The first click marks the upper-left vertex of the rectangle and the second click marks the lower-right vertex.

Bounding boxes can be:

Used together with tasks like Object/entity detection, Transcription, Classification, and Semantic segmentation.
Nested with Classification and Transcription tasks.

In Image projects, when you copy&paste an object (for example a bounding box), the contents of its sub-jobs, if any, are also copied. This can help speed up labeling similar complex objects.

For object-detection jobs with several classes and nested transcription sub-jobs, when you change the class by using the annotation context menu, the contents of the nested transcriptions will get transferred to the new class. Several limitations apply, though:

Instructions for nested jobs have to be the same between both classes
The types of nested jobs (including limits imposed on acceptable values, like data-only, or number-only entries) have to be the same between both classes

For PDF-related jobs:

It is not possible to make named entity recognition (NER) annotations where a bounding box has already been made. In this situation you can:

First make NER annotations, and then add the bounding box annotations above the exiting NER annotations.
Using the job viewer, hide specific bounding boxes and then annotate named entities.

Refer to these topics for information on how to use bounding boxes with: