Flexible implementation in PyTorch of the Variational Autoencoder (VAE), firstly introduced in:
Diederik P. Kingma and Max Welling. "Auto-encoding variational bayes." arXiv preprint arXiv:1312.6114 (2013). URL http://arxiv.org/abs/1312.6114.
The architecture of the VAE is customisable via command line, run train_vae.py --help for more details.
Constraint: the architecture of the decoder is the transposed of the encoder's.
The VAE with the default parameters was trained on the CelebA dataset. This pre-trained model is available in the models directory in this repository. The dataset is available at its official website. See some results below.
Training images (original vs. reconstruction)
Testing images (original vs. reconstruction)
New examples