A question about the DICELossMultiClass

[shreyaspadhy]

A question about the DICELossMultiClass

[HadiZayer]

I'm assuming that in a multi-class dice loss, you'd calculate the dice loss for each label that is not background and then average them. In this case, there is only one label that is not background, so it's hard-coded for the first label. Even though this sorts of make it not multi-class.

[xiaohu2015]

@HadiZayer Ok, I understand. You had better give a notice on the code.

[shreyaspadhy]

Yeah, @HadiZayer has it correct, it was a mistake on my part. Since I only used my UNet for two classes, I must have hard-coded that line during debugging. I'll make the necessary corrections. Thanks for bringing it to my attention!

[shreyaspadhy]

I've implemented multi-class DICE loss as of my last push.