ENH: Backend Annotation Restructuring

[czaloom]

Feature Type

• Adding new functionality to valor

• Changing existing functionality in valor

• Removing existing functionality in valor

Problem Description

Query solving is complex as a result of a cycle that is created in the table graph by the ground truth and prediction tables.

Feature Description

By doing the following we simplify the table graph and enable better support for generating queries.

• Remove GroundTruth Table

• Remove Prediction Table

• Add geometry linker table (similar to how embedding is implemented)

• Add metadata linker table

• Update annotation table

  • Add optional score
  • Add label id
  • Add geometry_id
  • Add metadata_id
  • Remove bounding_box, polygon, multipolygon, raster
  • Remove metadata

Predictions are distinguished from Ground Truths by the presence of a model definition.

Additional Context

This is setting the stage for nested queries that will allow complex user requests such as.

evaluation_filter = [
  GroundTruth.labels.in_([Label("class", "swimmer")]),
  Datum.where(GroundTruth.labels.in_([Label("class", "boat")])),
]

This is most easily accomplished when annotation, groundtruth and prediction are all represented the same table as they already share identical relationships.