Dice loss

转自：咖啡味儿的咖啡
https://blog.youkuaiyun.com/wangdongwei0/article/details/84576044

Dice Loss

首先定义两个轮廓区域的相似程度，用A、B表示两个轮廓区域所包含的点集，定义为：

                                                            

那么loss为：

可以看出，Dice Loss其实也可以分为两个部分，一个是前景的loss，一个是物体的loss，但是在实现中，我们往往只关心物体的loss，Keras的实现如下：

 
 

   
   

    
    

   
   
   
   

    
    

     
     def dice_coef(y_true, y_pred, smooth=1):
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
         intersection = K.sum(y_true * y_pred, axis=[
     
     1,
     
     2,
     
     3])
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
         union = K.sum(y_true, axis=[
     
     1,
     
     2,
     
     3]) + K.sum(y_pred, axis=[
     
     1,
     
     2,
     
     3])
    
    
   
   

   
   

    
    

   
   
   
   

    
    
    
     
     return K.mean( (
     
     2. * intersection + smooth) / (union + smooth), axis=
     
     0)
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
     def dice_coef_loss(y_true, y_pred):
    
    
   
   

   
   

    
    

   
   
   
   

    
    
	
     
     1 - dice_coef(y_true, y_pred, smooth=
     
     1)
    
    
   
   
 
 

上面的代码只是实现了2分类的dice loss，那么多分类的dice loss又应该是什么样的呢？

注：正确性还没验证。。。

 
 

   
   

    
    

   
   
   
   

    
    

     
     # y_true and y_pred should be one-hot
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
     # y_true.shape = (None,Width,Height,Channel)
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
     # y_pred.shape = (None,Width,Height,Channel)
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
     def dice_coef(y_true, y_pred, smooth=1):
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
         mean_loss = 
     
     0;
    
    
   
   

   
   

    
    

   
   
   
   

    
    
    
     
     for i 
     
     in range(y_pred.shape(
     
     -1)):
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
             intersection = K.sum(y_true[:,:,:,i] * y_pred[:,:,:,i], axis=[
     
     1,
     
     2,
     
     3])
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
             union = K.sum(y_true[:,:,:,i], axis=[
     
     1,
     
     2,
     
     3]) + K.sum(y_pred[:,:,:,i], axis=[
     
     1,
     
     2,
     
     3])
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
         mean_loss += (
     
     2. * intersection + smooth) / (union + smooth)
    
    
   
   

   
   

    
    

   
   
   
   

    
    
    
     
     return K.mean(mean_loss, axis=
     
     0)
    
    
   
   

   
   

    
    

   
   
   
   

    
    

     
     def dice_coef_loss(y_true, y_pred):
    
    
   
   

   
   

    
    

   
   
   
   

    
    
	
     
     1 - dice_coef(y_true, y_pred, smooth=
     
     1)