Background as a class in Semantic Segmentation

[Aleksandar1932]

I am doing binary semantic segmentation where I have one class that needs to be detected and the rest is background. Should background be considered as a class, hence setting the MODEL.SEM_SEG_HEAD.NUM_CLASSES = 2, or it should be set to 1.

The training is done on a custom dataset which is registered according to the detectron2 documentation, and each sem_seg_file is a binary mask.