I made some minor changes in order to make it a bit clearer.

[jpcosec]

In model.py

• Changed the object name to BVae instead of Vae
• Changed how some layers and the loss were declared in order to make it graph readable in tensorboard.
• Changed names to encoder, decoder, etc.
• Separated Z network method from the encoder and the decoder.
• Introduced histograms for the Z variable.

In main.py

• Changed Vae call to Bvae.

Pd: Nice work, it works just fine.