pytorch-dice-loss

Re-implementation of Dice Loss for NLP Tasks:

• Added more comments
• Simplified some computation

Installation

pip install pytorch-dice-loss

Usage

from pytorch_dice_loss import DiceLoss

loss = DiceLoss(with_logits=False, reduction='mean')
# [B, S, C]
input = torch.FloatTensor(
    [[[1, 0, 0], [0, 0, 1], [0, 0, 1], [0, 1, 0], [1, 0, 0]]]
)
input.requires_grad = True
# [B, S]
target = torch.LongTensor([[0, 2, 2, 1, 0]])
mask=torch.BoolTensor([[True, True, True, True, True]]
output = loss(
    input, target, mask=mask)
)