How does CARLA determine types for semantic segmentation

[0graph]

So I have an FBX + XODR and when they are imported into CARLA (using --no-carla-materials) they generate 4 folders Road, RoadLine, Sidewalk, Terrain

From what I understand Tagger.cpp uses the folder name to determine the tag for each object which is then used by the semantic segmentation sensor.

What I noticed though was that my segmentation had more classes then just those 4. It identified Poles, TrafficLights, TrafficSigns, etc. So where is this extra logic happening?

Or is it for some reason using CARLA materials anyways even though I had the flag and since those are in the correct folders it works? And if so, why is it happening, does it do this because it just auto generates assets based on the OpenDRIVE?

Thanks!