Modifying the number of inferred labels (reducing labels for inference vs training)

[ParsaBan]

Hello,
I want to reduce the number of labels I'm using during inference. For example, if my dataset.json contains 12 labels, I'd like to have nnUNet_predict only use 5 out of those 12 labels. I know when using CNN, the user can pass a config file that tells the inference script what labels to preserve and which to discard. Is a similar functionality present with nnUNet?

I'd ideally like to be able to set any labels not in a list to have a value of 0 after inference, or alternatively, have inference skip over those uncalled labels entirely. Any help would be appreciated. Thank you.

EDIT: Solved it by writing a script that loads the output generated by inference and sets "unneeded" labels to a value of 0 (background image)