DiCyR: Disentangled cyclic reconstruction for domain adaptation

Official github repository for the paper Disentangled cyclic reconstruction for domain adaptation submited to ICLR 2021. This repository contains the code and notebooks illustrating the experiments presented in the paper.

Installation

First clone the repository:

git clone https://github.com/AnonymousDiCyR/DiCyR.git
cd DiCyR

Create a virtual environment:

conda env create -f environment.yaml

Create the data folder:

mkdir data

Download the 3D shapes data here and copy it into the data folder.

Notebooks:

All notebooks are available here Each of them represents one experiment iteration.

Pre-trained models:

Pre-trained models are availables here