GANTheftAuto is a fork of Nvidia's GameGAN, aiming to emulate complex game environments, specifically Grand Theft Auto 5 (GTA5). It targets researchers and developers interested in generative adversarial networks for game simulation, offering improvements over the original GameGAN for more flexible and advanced experimentation.

How It Works

This project utilizes Generative Adversarial Networks (GANs) to generate realistic game environments. It builds upon Nvidia's GameGAN architecture but introduces significant enhancements, including support for PyTorch 1.8.1, non-square image generation (16:8 aspect ratio), larger generator/discriminator models, and the ability to use multiple generators. A key advantage is the inclusion of an inference script and live training previews, facilitating easier experimentation and evaluation of generated outputs.

Quick Start & Requirements

• Install: Clone the repository, install PyTorch with CUDA support (ROCm/CPU not supported), then run pip3 install -r requirements.txt.
• Prerequisites: CUDA-capable NVIDIA GPU (8GB VRAM minimum for training, 16GB+ recommended for larger batch sizes), Python.
• Demo: Execute ./scripts/gtav_inference_demo.sh or scripts\gtav_inference_demo.bat.
• Training: Use ./scripts/gtav_multi_demo.sh or scripts\gtav_inference_demo.bat.
• Docs: https://github.com/Sentdex/GANTheftAuto

Highlighted Details

• Fork of Nvidia's GameGAN with focus on GTA5.
• Supports PyTorch 1.8.1, non-square images, larger models, and multiple generators.
• Includes inference scripts and live training previews.
• Offers data collection scripts for GTA5, Vroom, and Cartpole environments.
• Provides sample datasets for training.

Maintenance & Community

The project is maintained by Sentdex. Community interaction and contributions are encouraged. Further details on community channels or roadmaps are not explicitly provided in the README.

Licensing & Compatibility

The repository's licensing is not explicitly stated in the provided README text. Compatibility for commercial use or closed-source linking would require clarification of the license.

Limitations & Caveats

The project is described as "still in progress," with acknowledged room for improvement. The Cartpole environment has a reported issue affecting action alternation. Inference functionality is noted as unfinished, with randomly generated actions. Data collection scripts for GTA5 are not shared, relying on provided sample datasets.