#stuff-and-things

Panoptic segmentation splits classes into countable "things" and amorphous "stuff". In a street-scene dataset, which of these classes would normally be stuff? Select all that apply.