Variational-autoencoder

A tensorflow implementation of a variational autoencoder.

Requirements:

1. tensorflow v1.13.1
2. PIL
3. matplotlib
4. IPython.display (for tensorboard visualization in a jupyter notebook)
5. OpenCV (for face cropping)

Usage:

Run the enclosed vae_mnist.ipynb notebook in a tensorflow environment.

Tensorflow graph:

From the enclosed vae_mnist.ipynb notebook, you can run tensorboard to see the entire computational graph.

A look at the entire tensorflow graph: 

A look at the variational autoencoder graph: 

A look at the encoder graph (three-layer fully-connected): 

A look at the decoder graph (three-layer fully-connected):