Is your feature request related to a problem? Please describe.
Feature request from @holgerroth:
I'm looking for something similar to the question asked here https://discuss.pytorch.org/t/make-one-hot-encoding-with-ignore-label-for-semantic-segmentation/18126/10
The background can be thought of as a ignored label. We could even have a list of labels to ignore when converting to one-hot, I think.
Predicting the background in the network is basically redundant in this case.