Get full object semantics

[LambdaGuard]

In current environment, obs.semantic while using GROUND_TRUTH_SEMANTIC=TRUE provides the semantic of goal object and all receptacles. Is there a way to get the full semantic of all objects inside the scene (just like the segmentation in obs.task_observations["instance_map"])?

[LambdaGuard]

Or, can we get the mapping between head_panoptic ids and their categories?

[yvsriram]

Hi! Yes, that's right. It provides semantic of goal object and all receptacles.

You can use these changes: https://github.com/facebookresearch/habitat-lab/compare/12e735d2d895b7878712e1ef64b7d5f7eeb997f3..b00978fc8113c7d56b8240665f0b0367ecf4eb0f for getting segmentation of all objects in scene.

The _cache_objects function contains code for mapping panoptic ids to categories.