Gaussian Splatting MLX Implementation

This repository contains a training implementation of 3D Gaussian Splatting using Apple's MLX framework.

Overview

Gaussian Splatting is a novel approach to Neural Radiance Fields (NeRF) that offers real-time rendering capabilities while maintaining high visual fidelity. This implementation leverages MLX - Apple's efficient machine learning framework designed specifically for Apple Silicon.

References

This implementation is based on:

• gaussian-splatting by INRIA Graphics Lab
• torch-splatting

Implemented Features

• Complete training pipeline for 3D Gaussians
• Chart of train loss
• PLY export for intermediate results visualization
• Support for COLMAP formatted datasets
• Support for NerfStudio formatted datasets
• Gaussian pruning and splitting operations
• Checkpoint visualization

Upcoming Features

• Memory optimization
• Use ARKit depth as dataset

Requirements

• Compatible with:
  • iPhone
  • iPad
  • Mac (runs as iPad-compatible app)
• MLX framework

ScreenShots

Citation

If you find this implementation useful, please cite the original Gaussian Splatting paper:

@article{kerbl3Dgaussians,
  title={3D Gaussian Splatting for Real-Time Radiance Field Rendering},
  author={Kerbl, Bernhard and Kopanas, Georgios and Leimk{\"u}hler, Thomas and Drettakis, George},
  journal={ACM Transactions on Graphics},
  volume={42},
  number={4},
  year={2023},
  publisher={ACM New York, NY, USA}
}