vae.autoencoder

Classes

`KLDivergenceLayer`	Custom layer to compute and add KL divergence to the model's loss.
`VAE`	VAE represents a Deep Convolutional variational autoencoder architecture

Functions

sample_point_from_normal_distribution(args)

Sampling function using the reparameterization trick.

Module Contents

vae.autoencoder.sample_point_from_normal_distribution(args): Sampling function using the reparameterization trick.

class vae.autoencoder.KLDivergenceLayer

Bases: tensorflow.keras.layers.Layer

Custom layer to compute and add KL divergence to the model’s loss.

call(inputs)

class vae.autoencoder.VAE(input_shape, conv_filters, conv_kernels, conv_strides, latent_space_dim)

VAE represents a Deep Convolutional variational autoencoder architecture with mirrored encoder and decoder components.

input_shape

conv_filters

conv_kernels

conv_strides

latent_space_dim

reconstruction_loss_weight = 10000

encoder = None

decoder = None

model = None

_num_conv_layers

_shape_before_bottleneck = None

_model_input = None

summary()

compile(learning_rate=0.0001)

train(x_train, batch_size, num_epochs)

save(save_folder='.')

load_weights(weights_path)

reconstruct(images)

classmethod load(save_folder='.')

_calculate_combined_loss(y_target, y_predicted)

_calculate_reconstruction_loss(y_target, y_predicted)

_create_folder_if_it_doesnt_exist(folder)

_save_parameters(save_folder)

_save_weights(save_folder)

_build()

_build_autoencoder()

_build_decoder()

_add_decoder_input()

_add_dense_layer(decoder_input)

_add_reshape_layer(dense_layer)

_add_conv_transpose_layers(x)

_add_conv_transpose_layer(layer_index, x)

_add_decoder_output(x)

_build_encoder()

_add_encoder_input()

_add_conv_layers(encoder_input)

_add_conv_layer(layer_index, x)

_add_bottleneck(x)