An implementation of the models presented in the Multi-View Data Generation Without View Supervision by Mickael Chen, Ludovic Denoyer and Thierry Artières, ICLR 2018

We propose a generative models for multi-view data by decomposing the latent space between content and view.

The code runs using PyTorch and numpy.

Each file is a stand-alone for the training of one model.

python gmv.py

gmv and cgmv are proposed model. gan2 is a simple baseline described in the paper. mathieu is a pytorch reimplementation of Disentangling factors of variation in deep representations using adversarial training using DCGAN inspired architecture.

Hyperparameters are set within the code and can be modified.