Skip to content

Bounding boxes for ML training data

At a glance

Draw bounding boxes around animals, humans, or vehicles in a classification, to produce training data for ML object-detection models.

  • Time: a few minutes per resource
  • Who: annotators producing training data
  • Prerequisites: a classified resource

Choosing an interface

There's only one place to draw bounding boxes — the Citizen Science frontend. (There's no separate "Expert" classification UI: the Django admin has no draw-on-image tooling at all.) Two routes get you to the same drawing view, depending on where you're starting from:

  1. Navigate to your project, open Classification view in the left sidebar.
  2. Pick the deployment you want to work on from the Deployment filter.
  3. Click a thumbnail's Classify link.
  4. Draw bounding boxes by clicking and dragging on the image, after Select new object.
  5. Associate each bounding box with the appropriate classification attributes, then click Classify to save.

Useful when you found the resource via Media (filtering/searching) rather than by deployment.

  1. On the Media list, click the preview icon on a row to open its detail view.

    Media detail view showing AI and user classifications

  2. Click the icon tooltipped "Classify image" (next to the AI/user classification tables) to open the drawing interface.

    Draw bboxes view

  3. Click Select new object, draw a box around the object as tightly as possible, then specify which dynamic attribute field it links to.

    Draw bboxes view with bboxes drawn

  4. Click Classify to save.

    Media detail view with bboxes drawn and classified

Verify it worked

Saved bounding boxes appear overlaid on the resource in the classification screen, linked to their classification attribute.

Next steps