Tensorflow Batchnormalization - 类型错误：轴必须是整数或列表，类型给定：<class 'tensorflow.python.framework.ops.Tensor'>

Question

以下是我训练 U-Net 的代码。 它主要是带有我自己的损失函数和指标的普通 Keras 代码，这对错误来说并不重要。 为了避免过度拟合，我尝试在每个卷积层之后添加一个 BatchNormalization 层，但是，我不断收到一个非常奇怪的错误。

inputs = tf.keras.layers.Input((self.height, self.width, self.channel))
c1 = tf.keras.layers.Conv2D(16, (3, 3), padding='same')(inputs)
c1 = tf.keras.layers.BatchNormalization(c1)
c1 = tf.keras.layers.LeakyReLU(self.alpha)(c1)
c1 = tf.keras.layers.Dropout(self.dropout_rate)(c1)
c1 = tf.keras.layers.Conv2D(16, (3, 3), padding='same')(c1)
c1 = tf.keras.layers.LeakyReLU(self.alpha)(c1)
c1 = tf.keras.layers.Dropout(self.dropout_rate)(c1)
p1 = tf.keras.layers.MaxPooling2D((2, 2))(c1)

....

u9 = tf.keras.layers.Conv2DTranspose(16, (2, 2), strides=(2, 2), padding='same')(c8)
u9 = tf.keras.layers.concatenate([u9, c1], axis=3)
c9 = tf.keras.layers.Conv2D(16, (3, 3), padding='same')(u9)
c9 = tf.keras.layers.LeakyReLU(self.alpha)(c9)
c9 = tf.keras.layers.Dropout(self.dropout_rate)(c9)
c9 = tf.keras.layers.Conv2D(16, (3, 3), padding='same')(c9)
c9 = tf.keras.layers.LeakyReLU(self.alpha)(c9)
c9 = tf.keras.layers.Dropout(self.dropout_rate)(c9)

outputs = tf.keras.layers.Conv2D(self.num_classes, (1, 1), activation='softmax')(c9)

self.model = tf.keras.Model(inputs=[inputs], outputs=[outputs])

self.model.compile(optimizer=tf.keras.optimizers.Adam(lr=self.learning_rate),
                   loss=cce_iou_coef,
                   metrics=[iou_coef, dice_coef])

每当我尝试添加 BatchNormalization 层时，都会出现以下错误。 我找不到问题，我做错了什么？

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-4-5c6c9c85bbcc> in <module>
----> 1 unet_dev = UNetDev()
      2 unet_dev.summary()

~/Desktop/notebook/bachelor-thesis/code/bachelorthesis/unet_dev.py in __init__(self, weight_url, width, height, channel, learning_rate, num_classes, alpha, dropout_rate)
     29             inputs = tf.keras.layers.Input((self.height, self.width, self.channel))
     30             c1 = tf.keras.layers.Conv2D(16, (3, 3), padding='same')(inputs)
---> 31             c1 = tf.keras.layers.BatchNormalization(c1)
     32             c1 = tf.keras.layers.LeakyReLU(self.alpha)(c1)
     33             c1 = tf.keras.layers.Dropout(self.dropout_rate)(c1)

~/anaconda3/envs/code/lib/python3.7/site-packages/tensorflow_core/python/keras/layers/normalization.py in __init__(self, axis, momentum, epsilon, center, scale, beta_initializer, gamma_initializer, moving_mean_initializer, moving_variance_initializer, beta_regularizer, gamma_regularizer, beta_constraint, gamma_constraint, renorm, renorm_clipping, renorm_momentum, fused, trainable, virtual_batch_size, adjustment, name, **kwargs)
    167     else:
    168       raise TypeError('axis must be int or list, type given: %s'
--> 169                       % type(axis))
    170     self.momentum = momentum
    171     self.epsilon = epsilon

TypeError: axis must be int or list, type given: <class 'tensorflow.python.framework.ops.Tensor'>

Answer 1

只需更换

c1 = tf.keras.layers.BatchNormalization(c1)

经过

c1 = tf.keras.layers.BatchNormalization()(c1)

就像 keras 中的其他层一样，这是调用它们的方式。 您正在向 keras 层提供参数，如文档中所示。 你不需要它