/gen-tsne

Primary LanguagePythonMIT LicenseMIT

Gen TSNE: Visualization and Metric for generative models

CI

Introduction

This repository contains the implementation of the valuation method proposed in the paper Demonstrating the Evolution of GANs through t-SNE.

See below an overview of the evaluation method:

We applied this method in COEGAN to provide further evidence of the evolutionary contribution of the model to the creation of strong generators and discriminators.

See below some results:

A metric based on the Jaccard index between t-SNE maps was designed to quantitatively represent the aspects of the model.

See below the results of the Jaccard index applied in experiments with COEGAN in the MNIST dataset:

Instructions

First, put images from the dataset and from generative models into different folders.

Then, start the process with the following command:

python main.py -b <DATASET IMAGES> -p <IMAGES FROM MODEL 1> -p <IMAGES FROM MODEL 2>

Execute python main.py --help to see more options.

If you want to use features instead of image pixels in the grid calculation (-f argument), the directory should follow the same structure used in test/assets/dataset and test/assets/model_a, i.e. store .npz (or .npy) files with the same name as each image that you want to evaluate.