This repo contains the code used to generate sprite datasets used to test disentangled models.

The code in adapted from the one found in the Spriteworld repo, which is used to generate simple reinforcement learning scenarios. It was also used to generate some of the datasets used to test the Spatial Broadcast Decoder. However, being designed for reinforcement learning tasks, it is not straigtforward to use it for static images and the authors do not provide their implementations. This code adapts the orignal one for that purpose, removing the unnecessary reinforcement learning code.

The reference for this repo is this one, should you decide to site it:

@misc{spritegen21,
author = {Milton Llera Montero},
title = {Spritegen: A flexible framework for dataset generation},
howpublished = {https://github.com/mmrl/spritegen/},
url = {https://github.com/mmrl/spritegen/},
year = "2022",
}

Additionally, the Spriteworld citation can be found here.