Labeling Interface

Let’s explore the complex example of multi-task labeling which includes text + image + audio data objects:

Key concepts

Here you can see relations among labeling objects: tasks, completions, results, etc.

One user provides one completion, it’s atomic, and it consists of the result items. my_projectResult items can have relations between themselves with the specified direction of three types: left-right, right-left, or bidirectional. Normalizations are additional information in the custom string format about the current result item.

Completions and Predictions are very similar. But predictions must be generated automatically by ML models.


Most of the actions described in this section are similar for all the data objects (images, audio, text, etc.).

Choices, TextArea and other simple tags

Such tags have straightforward labeling mechanics. It’s intuitive for users, so let’s talk about more complex things below.

Add region

  1. Select label you want to add (if you use Tag without labels like Polygon, just go to 2)
  2. Click on your data object (image, audio, text, etc)

Change label

You can change the label of the existing region:

  1. Select entity (span, bounding box, image segment, audio region, etc)
  2. Select a new label

Delete entity

  1. Select entity
  2. Press Backspace or go to Results panel and remove selected item

Add relation

You can create relations between two results with

  1. Select a first region (bounding box, text span, etc)
  2. Click on “Create Relation” button
  3. Select the second region
  4. Optionally: After the relation is created you can change the direction by click on the direction button
  5. Optionally: If you’ve configured labels, click on the triple dots button and add your predefined labels


Use hotkeys to improve your labeling performance. Hotkeys help is available in the labeling settings dialog.

ctrl+enterSubmit a task
ctrl+backspaceDelete all regions
escapeExit relation mode
backspaceDelete selected region
alt+shift+$nSelect a region