NVIDIA-Omniverse OVRTX semantic AOV capture
Use ovrtx-render-products-and-aovs for product/readback mechanics and
extend-ovrtx-semantic-aovs for implementation.
- Define durable semantic class/instance ownership in USD separately from the visualization. Preserve stable prim identity through composition.
- Add an explicit product/RenderVar path and typed native builder/decoder. Probe the installed native module's capabilities; fail as unsupported when the semantic builder or decoder is absent.
- Preserve dtype, dimensions, background/unlabeled value, class/instance mapping, sample time, product identity, and terminal status.
- Test duplicate labels, missing labels, hidden/excluded prims, remapped prim paths, iterator pages, wrong dtype/shape, terminal errors, and cleanup.
- Validate a tiny fixture with two known labeled objects and one unlabeled background before a real scene. Require exact pixel IDs at known regions and stable mapping across repeated reads.
Colorizing an ID array is a derivative for inspection. Keep the raw typed array and mapping authoritative. Do not claim support from a beauty image, material color substitution, or screen-space postprocess.