mfouda / max_sliced_wasserstein_distance

max_sliced_wasserstein_distance

Max Sliced-Wasserstein Autoencoder - PyTorch

Implementation of Max Sliced Wasserstein Distance in the paper "Generalized Sliced Wasserstein Distances" using PyTorch.

Declaration

This repo is based on the implementation shared by Emmanuel Fuentes, here I only modified the way of obtaining theta.

Requirement

To run this demo, please install the required packages by running: pip install -r requirements-dev.txt

Train the model with different setups

You can train this model with 'max' and 'normal' mode, which means using the Maximum Sliced-Wasserstein distance and the normal Sliced-Wasserstein distance, respectively.

To train with 'max' mode please run: python examples/mnist.py --mode 'max' --mode_test 'max'.

For more informations, please refer to this file: mnist.py