VAE-MNIST-Animator

A couple VAE (variational autoencoder) experiments I ran on the MNIST dataset in Keras, including animating the transition between digits in the latent encoding space.

Experiments included:

1. Visualizing the VAE Latent Space (stolen from Keras sample script)
2. Comparing reconstructed images to original images (stolen from Keras sample script)
3. Generating new images from arbitrary encodings
4. Animating transition between digits in the latent vector space

Sample Animation (3 to 0):

Built With:

• Python 3.6 (via Anaconda)
• Keras 2.0.8 (Tensorflow backend)
• Matplotlib 2.1.0

Sources:

• https://blog.keras.io/building-autoencoders-in-keras.html
• https://github.com/fchollet/keras/blob/master/examples/variational_autoencoder.py