Neural Radiance Fields - Flyte Tutorial

This repository contains a basic tutorial on using Flyte to train Neural Radiance Field (NeRF) models to compress basic bitmap images.

Table of Contents

• Introduction
• Setup
• Usage

Introduction

Neural Radiance Fields (NeRF) is a method for representing 3D scenes using neural networks. This tutorial demonstrates how to use Flyte to orchestrate and manage the training and evaluation of a NeRF model for basic 2D images.

Setup

To get started, follow these steps:

1. Clone the repository:

   git clone https://github.com/granthamtaylor/nerf

2. Install the required dependencies:

   • uv
   • just

3 Initialiize python environment

```sh
uv venv
uv sync
```

Usage

This tutorial includes several Flyte tasks and workflows to train and evaluate a NeRF model. Here are the basic steps to run the tutorial:

1. Define Flyte tasks:

   Flyte tasks are defined in the tasks directory. Each task represents a unit of work, such as data preprocessing, model training, or evaluation.

2. Define Flyte workflows:

   Workflows are defined in the workflows directory. A workflow orchestrates multiple tasks to achieve a specific goal, such as training a NeRF model.

3. Run the workflow:

   Use the Flyte CLI or Flyte console to launch the workflow. For example:

   • Run the model training workflow with a small image locally: just dev
   • Run the model training workflow with a larger image remotely just run