Semantic Labels#
Overview#
This capability provides requirements for semantic label attributes. These semantics may be used to provide ground truth for ML training by verifying perception system identification and classification of objects
Summary#
Geometry is semantically labeled
Semantic labels can be inherited from ancestral prims as well as bound materials
NVIDIA uses the QCode taxonomy for its Omniverse Asset libraries. Assets intended to be used alongside NVIDIA assets should use the same taxonomy.
Granularity#
Every renderable mesh must resolve a semantic label — this is the requirement, SL.001. Because labels are inherited, a single label on a common ancestor (typically the asset’s defaultPrim / root) satisfies it for all descendant geometry, so one label per asset is the minimum — as long as every renderable prim sits under a labeled ancestor, carries its own label, or is labeled through its bound material. Any renderable geometry outside a labeled subtree must be labeled separately.
In practice, label the asset as a whole first (“car,” “pedestrian,” “forklift”). Additional labels on parts within the asset (e.g. “tire,” “windshield,” “hubcap”) are optional and increase the usefulness of the semantic labels for ML training.
Taxonomy#
A taxonomy is identified by the SemanticsLabelsAPI instance name (for example, SemanticsLabelsAPI:class). This capability is taxonomy-agnostic: any instance name is valid, and thanks to the OpenUSD SemanticsLabelsAPI “Multiple Apply” schema design, multiple taxonomies can coexist on the same prim without conflict.
NVIDIA’s Omniverse Asset libraries use the Wikidata Q-code taxonomy (instance name wikidata_qcode). Assets intended for use alongside NVIDIA assets should use the same taxonomy; its formatting and validity rules are an NVIDIA-specific requirement — see Semantic Label QCode Valid (SL.QCODE.001).
Schema / OpenUSD Specification#
Semantic Labels are defined with the SemanticsLabelsAPI schema. Available in OpenUSD 24.11 and later.
USDA Sample#
The instance name (here class) identifies the taxonomy; any taxonomy may be used.
def Xform "Vehicle" (
prepend apiSchemas = ["SemanticsLabelsAPI:class"]
)
{
token[] semantics:labels:class = ["car"]
}
Requirements#
Tags |
Summary |
Compatibility |
Validator |
|---|---|---|---|
🔑 |
core-usd |
||
✅ |
Deprecated SemanticsAPI labels must be migrated to the SemanticsLabelsAPI schema. |
open-usd |
|
✅ |
open-usd |
||
✅ |
Material prims must be labeled with a non-empty |
omniverse |
|
✅ |
omniverse |
||
⛔ |
rtx |