Neural Network Projects with Python

Basic convolutional autoencoder

We're now ready to tackle the problem. Let's start with a basic model to see how far we can go with it.

As always, we define a new Sequential class:

basic_conv_autoencoder = Sequential()

Next, we add a single convolutional layer as our encoder layer:

basic_conv_autoencoder.add(Conv2D(filters=8, kernel_size=(3,3),                                  activation='relu', padding='same',                                   input_shape=(420,540,1)))

We add a single convolutional layer as our decoder layer:

basic_conv_autoencoder.add(Conv2D(filters=8, kernel_size=(3,3),                                   activation='relu', padding='same'))

Finally, we add an output layer:

basic_conv_autoencoder.add(Conv2D(filters=1, kernel_size=(3,3),                                   activation='sigmoid', padding='same'))

Let's check the structure of the model:

basic_conv_autoencoder.summary() ...