语义分割损失函数系列(3):Dice损失

Dice损失在医学图像分割任务中使用得极多，用于度量两个集合得相似性，dice系数的定义如下：
$$DiceCoefficient=\frac{2|A\cap B|}{|A|+|B|}$$

Dice 损失等于1-dice系数

$$Diceloss=1-DiceCoefficient$$

Pytorch实现的dice损失代码如下：

'''
Author: weifeng liu
Date: 2022-04-13 22:44:33
LastEditTime: 2022-04-14 14:29:34
LastEditors: Please set LastEditors
Description: 打开koroFileHeader查看配置 进行设置: https://github.com/OBKoro1/koro1FileHeader/wiki/%E9%85%8D%E7%BD%AE
FilePath: /BIBM-project/segmentation_pipeline/loss/dice.py
'''
import numpy
import torch
import torch.nn as nn
import torch.nn.functional as F

class DiceLoss(nn.Module):
    def __init__(self, weight=None, size_average=True):
        super(DiceLoss, self).__init__()

    def forward(self, inputs, targets, smooth=1):
        """
        Args:
            inputs (tensor): model outputs
            targets (tensor): image labels
            smooth (int, optional): smooth factor. Defaults to 1.

        Returns:
            loss 
        """
        #comment out if your model contains a sigmoid or equivalent activation layer
        inputs = F.sigmoid(inputs)       
        
        #flatten label and prediction tensors
        inputs = inputs.view(-1)
        targets = targets.view(-1)
        
        intersection = (inputs * targets).sum()                            
        dice = (2.*intersection + smooth)/(inputs.sum() + targets.sum() + smooth)  
        
        return 1 - dice