Semantic Label Setup

This page is for version Dev. For other versions, use one of these documentation indexes: - 3.3.0 (default): https://docs.nvidia.com/vss/3.3.0/llms.txt - Dev: https://docs.nvidia.com/vss/dev/llms.txt

Label the relevant objects so Ground Truth (GT) data can be generated for your targets. If the asset scene is SimReady, you can reuse its existing semantic labels for SDG. Otherwise, follow the steps below to add semantic labels.

Load the USD scene in Isaac Sim and open Window > Script Editor. Paste the contents of tools/sdg-postprocessing/semantic_labeling/box_check.py into a Python tab and add import omni.usd at the top before running.

• Using boxes in the scene as an example, first find the prim paths of the boxes to label by searching with keywords such as box. See box_check.py for details.

• Set your output_file path.

• Run the script and review the output .txt file to confirm the identified boxes are correct.

  • If they are correct, uncomment the matching enable_semantics(prim, category, "class") and/or enable_semantics(prim, 'basket', "class") calls, then rerun the script.
  • If they are not correct, refine the keywords in the script to narrow the search, then review the updated .txt file.

• To remove all labels in the scene, run remove_label.py in the same Script Editor session (after box_check.py defines get_stage()), then repeat the steps above.