def __call__(self, input, target):
"""
:param input: 5D probability maps torch tensor (NxCxDxHxW)
:param target: 4D or 5D ground truth torch tensor. 4D (NxDxHxW) tensor will be expanded to 5D as one-hot
:return: Soft Dice Coefficient averaged over all channels/classes
"""
# Average across channels in order to get the final score
return torch.mean(compute_per_channel_dice(input, target, epsilon=self.epsilon, ignore_index=self.ignore_index))
def __call__(self, input, target):
# Average across channels in order to get the final score
return torch.mean(compute_per_channel_dice(input, target, epsilon=self.epsilon))
def __call__(self, input, target):
# Average across channels in order to get the final score
class_idx = torch.arange(input.shape[1]).to(input.device)
input = torch.argmax(input, axis=1)==class_idx[:,None,None,None,None]
input = input.transpose(1,0)
return torch.mean(compute_per_channel_dice(input, target, epsilon=self.epsilon))