Gaussian Splatting for NeRF

This project implements Gaussian Splatting for 3D scene reconstruction, designed to work as part of a larger Neural Radiance Fields (NeRF) pipeline. It provides functionality for training Gaussian models, rendering scenes, and serving results through a web interface.

System Overview

The Gaussian Splatting system consists of several key components:

1. Job Dispatcher: Manages the queue of training jobs and dispatches them to available GPUs.
2. Training Pipeline: Implements the core Gaussian Splatting algorithm for learning 3D scene representations.
3. Rendering Engine: Renders the learned 3D scenes from various viewpoints.
4. File Server: Provides HTTP access to the generated outputs.
5. Utility Functions: Includes various helper functions for data conversion, output generation, and more.

Key Features

• Multi-GPU support for concurrent job processing
• RabbitMQ integration for job queueing and management
• Dynamic configuration of training parameters
• Output generation in multiple formats (PLY, SPLAT, video)
• Web server for accessing rendered results

Getting Started

Prerequisites

• CUDA-capable 11.7 GPU(s)
• NVCC for Pytorch Extension compilation READ THIS
• Python 3.7+
• PyTorch
• RabbitMQ
• Flask

Installation

As a service (Docker) PREFERRED

Refer to the backend setup instructions

If you wish to manually build the container

1. Build the image

docker build -t nerf-worker  

2. Start the image

docker run nerf-worker

Locally

1. Clone this repository
2. Install the required Python packages:

   pip install -r requirements.txt
   

3. Set up RabbitMQ and configure the connection details in the .env file

Usage

1. Start the job dispatcher:

   python dispatcher.py
   

2. Start the file server:

   python fileserver.py
   

3. Submit jobs to the nerf-in queue in RabbitMQ

Configuration

The system can be configured through various parameters in the train.py file and environment variables. Key configuration options include:

• Number of training iterations
• Output types (PLY, SPLAT, video)
• GPU allocation strategy
• RabbitMQ connection details

Contributing

Contributions to improve the Gaussian Splatting system are welcome. Please submit pull requests or open issues for any bugs or feature requests.

License

This project is licensed under [LICENSE NAME] - see the LICENSE.md file for details.

Acknowledgments

• INRIA GRAPHDECO research group for the original Gaussian Splatting implementation
• [Spring '24 Team] for adapting and extending the system

For more detailed information on specific components and processes, please refer to the wiki pages.