ignore_label for one-class Semantic Segmentation

[Raneem-MT]

Hi. I was wondering how exactly to use ignore_label for a one-class semantic segmentation problem. I had binary masks (1 for class and 0 for background) and then did the pre-processing transformation you added and my values became 0 and 255, but the model is ignoring the 255 value during evaluation, and I am getting very low performance results.

I was wondering what you set ignore_label to and whether the model assumes the background by itself?
Thank you.