Neural Network Projects with Python

Deep convolutional autoencoder

Let's try denoising the images with a deeper model and more filters in each convolutional layer.

We start by defining a new Sequential class:

conv_autoencoder = Sequential()

Next, we add three convolutional layers as our encoder, with 32, 16, and 8 filters:

conv_autoencoder.add(Conv2D(filters=32, kernel_size=(3,3),                            input_shape=(420,540,1),                             activation='relu', padding='same'))conv_autoencoder.add(Conv2D(filters=16, kernel_size=(3,3),                            activation='relu', padding='same'))conv_autoencoder.add(Conv2D(filters=8, kernel_size=(3,3),                            activation='relu', padding='same'))

Similarly for the decoder, we add three convolutional layers with 8, 16, and 32 filters:

conv_autoencoder.add(Conv2D(filters=8, kernel_size=(3,3),  activation='relu', ...