GAN Pixel Art Swords

This project was part of a University AI module looking at machine learning. The module focused on using generative adversarial networks and convolutional neural networks to solve generative and classifier issues. The aim of this project was to create recognisable pixel art swords that could be used in a fantasy themed video game. 

Software: python, Google Colab 

Project aims:

• To create recognisable pixel art swords
• To create pixel art swords that could be utilised in pixel art video games
• To develop understanding of machine learning

Project development context:

• Part of optional AI module in final year at university
• Developed in Python
• Used machine learning to solve classifier and generative issues
• Focusing on CNNs (convolutional neural network) and GANs (generative adversarial network)
• Transfer learning was used to train the network
• Final project focus on GANs to support future game development projects without requiring ongoing access to bank of art designs

Project Reflection/ Evaluation:

• Aim to achieve recognisable pixel art swords was accomplished
• Pixel art swords generated resulted in good shape
• Standard achieved for several of the pixel art swords could have been utilised in video games
• Future development would involve a stronger training data set with more colour to create more colour variations in the images generated
• Future development work could include using the same technique to create other commonly occurring graphics in games

Colab Link:
https://colab.research.google.com/drive/12WRuXrZ8lNZP0XLg72vSkhAaxqAtsIhY?usp=sh...