Variational Autoencoder (VAE)

VAE is proposed by Kingma and Welling in 2014 on ICLR.
This code has implemented with pytorch version 1.0 and python3.
If you execute this source code, you type the command as shown below on terminal.

python3 main.py

This VAE can choice network type (i.e. FC only or CNNs).
Default network is FC only.
Usage is shown below:

• FC only: fully connected only

python3 main.py --network_type 'fc'

• CNNs: convolutional neural networks

python3 main.py --network_type 'cnn'

Result

Extra items

• Auto-Encoding Variational Bayes
• Tutorial on Variational Autoencoders
• VAE's Slide
• Chainer official implementation