Supported Annotations Types

Most of the highest leverage workflows in Nucleus are based on visualizing and querying on ground truth annotations, and evaluating your model predictions against them.

Task Types

Nucleus supports a variety of task types, namely:

• Geometric
  • Bounding Box
  • Polygon
  • Line
  • Cuboid
• Segmentation
• Classification

If you already have a labeling project, you can upload tasks to Nucleus directly and have their labels synced automatically! See Ingest From Labeling.

Working with the SDK

Nucleus uses a generalized, unopinionated payload schema for annotations. The recommended way to construct these payloads is using our Python SDK.

When working in the Python SDK, each Annotation object houses:

• Intrinsic information about the annotation itself (e.g. label, geometry).
• The reference ID of the item with which to associate the annotation.
• An optional unique annotation ID (if not provided, will be generated automatically).
• Optional metadata (stored as arbitrary key-value pairs).