Label 0 for MaskFormer Semantic Segmentation- Custom dataset

Hi everyone,

I’m fine-tuning MaskFormer for semantic segmentation to identify diseased areas on leaves, following [this tutorial] from Niels. My masks are binary (array([0, 1], dtype=uint8)) with id2label as {0: 'unlabeled', 1: 'diseased'}. Since my focus is solely on segmenting the diseased areas, I’m questioning whether including label 0 for background pixels are necessary or if I should revise my labeling approach. Any advice on the best practice for using label 0 in such a focused segmentation task would be greatly appreciated.