Properties (archived)

Enhance data labeling and classification in V7 with Properties: your solution for comprehensive tagging, categorization, and metadata enrichment in annotations

Overview

Properties enable users to add and edit detailed information to their annotations. This feature streamlines the labeling process and enriches your dataset for downstream ML development.

Properties allow you to ask specific questions about each annotation. This interactive approach enables the collection of intricate details that go beyond basic labeling, thereby creating a richly annotated dataset. The benefits of this feature are multifold: it fosters a deeper understanding of the annotated objects, enhances the quality of the training data for ML models, and allows for more nuanced AI behaviours based on these detailed annotations.

Property Types

Depending on what your use case requires properties can be either:

  • Single Select
  • Multi-Select

Creating Properties

Properties can be seamlessly added to both new and existing Classes via the V7 Class menu. The process involves naming the property, framing a guiding question for labelers, and specifying the options.

Note: Labelers are restricted from adding new property options during the labeling process; these must be defined in advance.

Annotating with Properties

Labelers can choose the appropriate properties for each annotation instance during labeling.

Property Editing

  • Select an annotation and use the right-hand toolbar in the workview.
  • Click on the centroid of an annotation to open the property editor.

📘

Viewing Properties

When working in the annotation interface, you can view the properties in the annotation side panel by using the "Toggle Sub-Annotations" button. If you prefer keyboard shortcuts for efficiency, simply press SHIFT-H to toggle the view of these properties.

Importing Properties

When you download any export containing at least one property, you'll notice that the export contains a metadata.json file inside a hidden .v7 folder. This file contains various metadata about each property present in the export including:

  • The property name
  • Whether it's single-select or multi-select
  • It's description
  • Whether it's a required property or not
  • The property values themselves along with their colours

For example:

"properties": [
        {
          "name": "single1",
          "type": "single_select",
          "description": "Property description",
          "required": false,
          "property_values": [
            {
              "value": "123",
              "color": "rgba(219,255,0,1.0)"
            },
            {
              "value": "12",
              "color": "rgba(255,0,122,1.0)"
            },
            {
              "value": "1",
              "color": "rgba(219,0,255,1.0)"
            }
          ]
        }
  ]

It's included because without it, if you were to import properties that don't yet exist on the relevant classes, the definitions in metadata.json are used to create & link those properties to the relevant classes.

Therefore, when importing properties that don't yet exist on the relevant classes, you'll need to either:

  • 1: Construct a .v7/metadata.json file, or:
  • 2: Pre-create and assign all properties to their classes before importing. This can be done manually in the UI or with the API. In this case, you will not need to create a .v7/metadata.json file.

Deleting Properties

Deleting properties from a class can be performed within the Classes menu. Here's what you need to know:

❗️

Considerations Before Deletion

Data Impact: Be aware that deleting a property will remove all associated annotations of this property across all datasets. This action is irreversible!

Steps to Delete a Property

  1. Navigate to the Classes menu.
  2. Locate the class from which you want to remove a property.
  3. Select the property to be deleted.
  4. Confirm the deletion. Ensure you understand the implications of this action.

🚧

What's Next?

Required properties are not part of this initial release and will be available in early 2024

Please send any feedback on properties to [email protected] or through your CSM! We are exploring different options to take properties to the next level, and we would love to hear what you think.