Did the soft_dice loss at "SpaceNet_Optimized_Routing_Solutions/xd_xd/aa/pytorch/loss/__init__.py /" work for multiclass?

[akhilnegi]

I believe the soft dice loss implementation at "SpaceNet_Optimized_Routing_Solutions/xd_xd/aa/pytorch/loss/init.py /"
can only work for binary classification. Since, spacenet 5 was a multiclass classification. What was the dice loss implementation used as mentioned by xd_xd "Loss=dice+focal."?