[FIXED] How can I get summary of customed tensorflow Model

Issue

How can I get the summary of customed tensorflow Model?

class Discriminator_block(tf.keras.layers.Layer):
  def __init__(self, num_strides):
    super(Discriminator_block, self).__init__(name='discriminator block')
    self.num_strides = num_strides
    self.conv1 = tf.keras.layers.Conv2D(filters=128, kernel_size=(3,3), strides=(num_strides, num_strides), padding='same', data_format='channels_first', activation=None)
    self.bn1 = tf.keras.layers.BatchNormalization(axis=1)
    self.leaky = keras.layers.advanced_activations.LeakyReLU()

  def forward(self, x):
    x = self.conv1(x)
    x = self.bn1(x)
    x = self.leaky(x)
    return x

I have coded my own discriminator blocks by using tensorflow and I want to see my models' summary

So I added

Discriminator_block.summary()

but I get a error:

'Discriminator_block' object has no attribute 'summary'

Which mistake did I make in my code?

Solution

You need to change and consider multiple steps in your code to print a summary of your custom block.

1. Important thing is to change tf.keras.layers.Layer to tf.keras.Model
2. Create an instance of your model.
3. Input a random tensor to the model that your model is building then you can get the summary of the model.

import tensorflow as tf
class Discriminator_block(tf.keras.Model):
    def __init__(self, num_strides):
        super(Discriminator_block, self).__init__(name='discriminator block')
        self.num_strides = num_strides
        self.conv1 = tf.keras.layers.Conv2D(filters=16, kernel_size=(3,3), 
                                            strides=(num_strides, num_strides), 
                                            padding='same', 
                                            activation='relu', 
                                            input_shape=(28,28,3))
        self.bn1 = tf.keras.layers.BatchNormalization(axis=1)
        self.leaky = tf.keras.layers.LeakyReLU()
        
    def call(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.leaky(x)
        return x
    
block = Discriminator_block(num_strides = 1)
_ = block(tf.random.normal(shape=[2, 28, 28, 3]))
block.summary()

---

Model: "discriminator block"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_1 (Conv2D)           multiple                  448       
                                                                 
 batch_normalization_1 (Batc  multiple                 112       
 hNormalization)                                                 
                                                                 
 leaky_re_lu_1 (LeakyReLU)   multiple                  0         
                                                                 
=================================================================
Total params: 560
Trainable params: 504
Non-trainable params: 56
_________________________________________________________________

Answered By - I'mahdi