Feature: ID:011 - Semantic Labels - Base#

Description#

This feature describes the requirements necessary to have semantic label attributes on geometry prims. These semantics may be used to provide ground truth for ML training by verifying perception system identification and classification of objects.

Property

Value

Proprietary Techs

Core USD, Open USD, NVIDIA Omniverse, RTX

Dependency

None

Neutral Format#

Version 0.2.0#

Details

Property

Value

Internal ID

FET011_BASE_NEUTRAL

Used in Profiles#

This version is not currently used in any profiles.

Requirements#

Used in Profiles#

This version is not currently used in any profiles.

Pipelines Supported for this Feature#

Source file type:

  • .blend

    • Via Blender SimReady Add-ons

  • .mjcf

    • Via Blender SimReady Add-ons + MJCF2USD Tool

  • .step

    • Via Blender SimReady Add-ons + CAD Converter

Test Process#

  • Obtain the usd sdk

  • Confirm your asset in question has passed validation

  • In your commandline type:

    • path/to/usdsdk/scripts/usdrecord <path to usdfile.usd> <path to output.png>

  • Open up path/to/output.png

  • Expected Result:

    • Confirm it is NOT empty or completely black

    • All geometry prims should have semantic labels

    • Semantic labels should use proper schema and not contain time samples

Comments#

This feature enables ML training ground truth by providing semantic label attributes on geometry prims. The requirements ensure that all geometry is properly labeled using the SemanticsLabelsAPI schema, with support for Wikidata ontology Q-Codes and proper time-invariant labeling for sensor simulation.