Skip to content

Latest commit

 

History

History
51 lines (34 loc) · 2.11 KB

README.md

File metadata and controls

51 lines (34 loc) · 2.11 KB
Raw

Generating 2D map tiles with Generative Adversarial Network (GAN)

Introduction

This repository contains all the source code used to generate small 2D tiles commonly used in video games for building maps. Generating tiles was achieved using a Deep Convolutional Generative Adversarial Network (DCGAN). This code uses PyTorch as backend.

This is an example of what it can generate:

And when interpolating in its latent space:

You can find more information and details about this source code in this article: https://playerone-studio.com/gan-2d-tiles-generative-adversarial-network

Dependencies

There are few dependencies:

How to use it

1. Create your own dataset:

  • You can use the simple Processing (https://processing.org/) script provided in this repository to convert downloaded tilesets into individual tiles
  • All tilesets must be in a folder with no other file, and tiles should be of the same size in all tilesets (here 32x32)
  • Tiles will be saved as individual PNG files. Empty tiles will be omitted.

Here are some samples of tiles I used for my dataset:

2. Train your model:

  • Edit train.py so that the paths match that of your dataset (images have to be power of 2)
  • Also adjust any settings as you want. The settings are detailed in the file gan.py
  • run using "python train.py"

You should be patient as etting the first results can take some time. Here are some results during training:

3. Test your model:

  • Edit test.py to match the path where you saved your model
  • Also adjust latent space dimension if required
  • run using "python test.py" to generate some test images in your output folder
  • images will be saved to the same path as your model

Enjoy!