/Wasserstein-Geodesic-Generator-for-Conditional-Distributions

Primary LanguagePython

Introduction

This supplementary file provides the implementation used for the results in Wasserstein Geodesic Generator for Conditional Distributions" [paper]

Results by the proposed method Results by the baseline method

Conditional generation results by the proposed method (left) and cAAE (right). The proposed method produces face images with clearer eyes, noses, and mouths than baselines. For each method, the leftmost and rightmost columns show generation results for observed domains and intermediate columns show results for unobserved intermediate domains.
A visualization of latent interpolation with the real data

A visualization of latent interpolation with the real data and their translation results by various methods. The bottom row shows real images from two observed domains. From the first to third rows, the leftmost column shows the ground-truth, the rightmost column shows transportation results of the ground-truth, and intermediate columns show latent interpolation results for unobserved intermediate domains.

Training Environment

Dependencies

python 3.6.5

cv2==3.4.3
logging==0.5.1.2
matplotlib==3.1.2
numpy==1.18.1
pandas==0.25.3
PIL==6.0.0
torch==1.0.1
torchvision==0.2.2
torchfile==0.1.0
tensorboardX==2.1
protobuf=3.13.0.1

anaconda virtual environment

We recommand to use anaconda virtual environment :

conda create -n venv python=3.6.5
conda activate venv
conda install pytorch==1.0.1 torchvision==0.2.2 cudatoolkit=10.0 -c pytorch

proper version of cudatoolkit is needed.

commands

to train the encoder, run:

python3 train_encoder.py

to train the translator, run:

python3 train_translator.py