Autoencoders safari

Convolutional Autoencoders implementations using tensorflow and keras and the MNIST dataset.

In this safari you will see three implementations of autoencoders: a convolutional autoenconder, a convolutional variational autoencoder and a convolutional conditional variational autoencoder.

Convolutional Autoencoder (CA)

convolutional_autoencoder.ipynb: implementation of a convolutional autoencoder.

• Both the encoder and the decoder have convolutional layers.
• You will also find some plots of the latent space.

Convolutional Variational Autoencoder (CVA)

variational_convolutional_autoencoder.ipynb: implementation of a convolutional variational autoencoder.

• We make use of the reparametrization trick.
• Both the encoder and the decoder have convolutional layers.
• You will also find some plots of the latent space.
• At the end you will find how to generate samples using the trained encoder.

Convolutional Conditional Variational Autoencoder (CCVA)

conditional_convolutional_variatinoal_autoencoder.ipynb: implementation of a convolutional conditional variational autoencoder.

• We make use of the reparametrization trick.
• Both the encoder and the decoder have convolutional layers.
• At the end you will find how to generate samples using the trained encoder and custom labels to select the category of the samples.