Snake Diffusion model

Overview

The Snake Diffusion project is an educational repository showcasing the implementation of a real-time Snake game using a diffusion model. Drawing inspiration from several significant research papers, this project offers a simplified approach to complex concepts in artificial intelligence and game development.

It utilizes insights from:

• Google's Doom Diffusion paper
• Oasis GitHub repository
• Diamond paper

The focus on the Snake game, due to its straightforward mechanics, enabled the development of an engaging proof of concept over a span of two months, which involved extensive experimentation to fine-tune the model for an enjoyable gameplay experience.

If GPU resources are not available, users have the alternative of utilizing runpod.io, a paid cloud service for computational needs.

Model Architecture

The chosen architecture for this implementation is the EDM diffusion model, recognized for its superior performance with fewer sampling steps compared to other models like DDIM, which necessitates a larger number of steps to achieve similar quality.

Gameplay Experience

Once the model is ready, users can enjoy playing the Snake game. The repository includes various options for both local execution and cloud-based gameplay through runpod.io:

1. For local play, running the provided Play.ipynb allows gameplay at a rate of 1 frame per second, influenced by GPU capabilities.
2. For cloud-based options, after deploying a suitable Pod (RTX 4090 recommended), users can execute the game scripts from runpod.sh and launch Play.ipynb for a seamless gaming experience.

This repository serves as a valuable resource for those interested in the intersection of AI and game development, providing insights into the practical application of diffusion models in an engaging format.