Beginner-Friendly
Advanced
Neural Radiance Fields (NeRFs) and Gaussian Splatting (Splats) are techniques for creating 3D representations from 2D images. They allow you to:
A lot of it boils down to creating photorealistic scenes from a bunch of images, that include reflections and other complex properties. Usually these properties are really difficult or sometimes impossible to capture with traditional photogrammetry tools.
Polycam has a fantastic page where they have an expert explain the difference in thorough detail.
1. Movement
2. Scene Selection
Nerfstudio is the biggest open-source project (that we know of) about anything NeRF related. They have thorough guides on how to train your first NeRF, how to set up your dependencies, and a lot of great general technical know-how on NeRFs.
The Google colab notebook that they provide is an amazing resource to get up and running with training your very first NeRF. Most of the code that we use below is taken from that notebook, and most days it works great. When it doesn't it's because COLMAP has given up, and that's when I recommend you install it locally and run it on your machine.
If you are running into issues with COLMAP - I suggest running a no-gpu flag on it, as it doesn't play well with cloud GPUs.
For some context, Google Colab, or Colaboratory, is a free, cloud-based tool that lets you write and run Python code in your browser.
This service is the easiest and fastest way for getting into NeRFs and Gaussian Splats. This is where I, Miro, started my adventure, and it has been a solid foundation for my curiosity.
Sadly, they have slowed the service down in the last few months, so expect to wait a bit before you get your first NeRF trained.
They have taken down their Best Practices guide, but I have made a copy of it here.
Promoted as 3D reconstruction for all, and we agree. This tool has come out only a few weeks ago from time of writing (Feb 2025), and it already does a lot of heavy lifting for local Gaussian Splat processing. It can export a .ply file that you can load up in After Effects or Unity, all on your machine at home. We like this tool, because it is the only one (so far) that works great with the M-chip MacBooks.
It is VERY early in development, currently on release 0.2.0, and in the next version there should be some more updates to how the Splats are rendered, and increasing their quality as well.
They have been around for a while and heavily invested in photogrammetry techniques. They heavily utilise the LiDAR on an iPhone to quickly scan and generate 3D models that can be used to create 3D models of objects and spaces.
They recently also introduced Gaussian Splats in their workflow, and have some robust editing and viewing tools for splats.