A tensorflow implementation of a variational autoencoder.
- tensorflow v1.13.1
- PIL
- matplotlib
- IPython.display (for tensorboard visualization in a jupyter notebook)
- OpenCV (for face cropping)
Run the enclosed vae_mnist.ipynb notebook in a tensorflow environment.
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):
