Keras：使用 Dice 系数损失函数，val loss 没有改善

Question

问题

我正在做两类图像分割，我想使用骰子系数的损失函数。但是验证损失没有得到改善。如何解决这些问题？

我做了什么

采用one-hot编码的方法，处理标签图像，并且没有包含背景标签。

代码

X 的形状是 (num of data, 256, 256, 1) # graysacle

y的形状是(num of data, 256, 256, 2) #二分类并排除背景标签

one_hot_y = np.zeros((len(y), image_height, image_width, 2))
for i in range(len(y)):
  one_hot = to_categorical(y[i])
  one_hot_y[i] = one_hot[:,:,1:] 
one_hot_y.shape  #->  (566, 256, 256, 2)

#### <-- Unet Model --> ####

from tensorflow import keras
from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, Concatenate, Conv2DTranspose
from keras import Model

def unet(image_height, image_width, num_classes):
    # inputs = Input(input_size)
    inputs = Input(shape=(image_height, image_width, 1),name='U-net')
    
    conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(inputs)
    conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv1)
    pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)

    conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(pool1)
    conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv2)
    pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)

    conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool2)
    conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv3)
    pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)

    conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool3)
    conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv4)
    pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)

    conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool4)
    conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv5)

    up6 = Concatenate()([Conv2DTranspose(256, (2, 2), strides=(2, 2), padding='same')(conv5), conv4])
    conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(up6)
    conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv6)

    up7 = Concatenate()([Conv2DTranspose(128, (2, 2), strides=(2, 2), padding='same')(conv6), conv3])
    conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(up7)
    conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv7)

    up8 = Concatenate()([Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(conv7), conv2])
    conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(up8)
    conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv8)

    up9 = Concatenate()([Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same')(conv8), conv1])
    conv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(up9)
    conv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv9)

    outputs = Conv2D(num_classes, (1, 1), activation='softmax')(conv9)
    
    return Model(inputs=[inputs], outputs=[outputs])```


#### <-- Dice Score --> ####

from tensorflow.keras import backend as K
def dice_coef(y_true, y_pred):
  y_true_f = K.flatten(y_true)
  y_pred_f = K.flatten(y_pred)
  intersection = K.sum(y_true_f * y_pred_f)
  return (2. * intersection + 0.0001) / (K.sum(y_true_f) + K.sum(y_pred_f) + 0.0001)

def dice_coef_loss(y_true, y_pred):
  return 1 - dice_coef(y_true, y_pred)```


#### <-- Fit the Model --> ####

from tensorflow.keras import optimizers
adam = optimizers.Adam(learning_rate=0.0001)
unet_model.compile(optimizer=adam, loss=[dice_coef_loss],metrics=[dice_coef])
hist = unet_model.fit(X_train,y_train, epochs=epochs, batch_size=batch_size,validation_data=(X_val,y_val), callbacks=[checkpoint,earlystopping])

Answer 1

我试图复制您的经验。我使用 Oxford-IIIT Pets 数据库，其标签分为三类：1：前景，2：背景，3：未分类。如果像您一样删除了第 1 类（“前景”），则 val_loss 在迭代期间不会更改。另一方面，如果“未分类”类被删除，优化似乎工作。该模型无法区分“背景”和“未分类”，这是可以想象的。
此外，骰子系数的计算有一个小误差：分母中需要取平方的和。它不会改变 y_true 的任何东西，但它会改变 y_pred。

Answer 2

我不能说为什么你的代码不起作用，但我可以告诉你我是怎么做的。不同之处在于我在骰子系数计算函数中排除了背景并编码了目标。

然后我定义我的骰子系数如下：

def dice_coef(y_true, y_pred, smooth=1):
    # flatten
    y_true_f = K.flatten(y_true)
    y_pred_f = K.flatten(y_pred)
    # one-hot encoding y with 3 labels : 0=background, 1=label1, 2=label2
    y_true_f = K.one_hot(K.cast(y_true_f, np.uint8), 3)
    y_pred_f = K.one_hot(K.cast(y_pred_f, np.uint8), 3)
    # calculate intersection and union exluding background using y[:,1:]
    intersection = K.sum(y_true_f[:,1:]* y_pred_f[:,1:], axis=[-1])
    union = K.sum(y_true_f[:,1:], axis=[-1]) + K.sum(y_pred_f[:,1:], axis=[-1])
    # apply dice formula
    dice = K.mean((2. * intersection + smooth)/(union + smooth), axis=0)
    return dice

def dice_loss(y_true, y_pred):
    return 1-dice_coef