Confused about "ignore mask" in the occ prediction task

[TabGuigui]

Hello, appreciate for your excellent work firstly.

In your occflow_label.py file and the previous works such as beverse , I have found that you all set a no-vehicle-object frame with a 255 segmentation gt and don't compute the loss in such frame.

I'm very confused the meaning of such operation，should not the model predict a all-zero segmenation result with a no-vehicle-object frame ?