Gaussian Splatting Research

Related concepts: Artificial Intelligence | Machine Learning | Computer Science | 3D Reconstruction

Okay summarize my research on affordable gaussian.

Hi Ann,

Here’s my research

I discovered a few pipelines that are affordable and relatively simple to use, It depends what you want and what for, if you want a high quality rendor, OpenSplat is free and opensource (big fan of FOSS software) and can run locally on cpu to convert images/videos from COLMAP files to get those points and recreate, you’d need a diff software for that. Postshot might be the best one atm since it’s very versatile, its got a lot in one. I dont have a nvidia graphics card so i cant run it on my pc. Ive used an alternative called x.

Ive experimented with Luma AI, Scaniverse and Polycam etc. And so far Scaniverse is the best cost-wise when it comes to easily scanning - its free and can export high quality .PLY files for further modification or postprocessing.

One of the best pipelines ive found to create an interactive digital twin is using Scaniverse -> Unreal Engine with the .luma plugin. You can add collisions to simulate walking etc.

XScene is so far the best for Unreal Engine since it has a rendering pipeline inbuilt, a bit finnicky but works. Its latest updates were in january this year so its a bit outdated but does the right job.

Depending on what engine your team intends to work with, i can probably tweak it based on that.

Preprocessing:

Scaniverse - Niantic -> Lets you use gaussian splatting for venues

Issue with a lot of Gaussian splatting apps like Luma AI is that they are localized to an object. Best for small things, but bigger like venues etc is always better to do scaniverse,

Post/Edit:

Supersplat:

Postshot: -> Full content ownership!! Very good. -> Free and in open beta. I cant use it so im using Brush.

Brush:

Uses Burn ML & can setup in Android studioa s well as online.

Note: Most are build on CUDA. So not a lot of AMD

OpenSplat -> Opoensource and runs from AMD

In summary, there is a huge range of options to create an interactive 3D digital twin of a venue on a budget. On the low-cost end, you can leverage open-source tools and a smartphone to scan and reconstruct the space (even using the latest Gaussian Splatting techniques for realism). Platforms like Luma AI and Polycam make it point-and-shoot simple to get a sharable 3D scene from imagesxraispotlight.substack.com, while nerfstudio and others let power-users dig into the process themselves. For adding interactive guidance, solutions like Matterport offer polished annotation featuresmatterport.com, and even free platforms like Sketchfab allow basic labels.

Depending on your exact needs (level of detail, interactivity, technical skill), you might mix and match these tools. For example, you could scan the stadium with Scaniverse to get quick splat models, refine them with Lichtfeld Studio or Meshroom if needed, and then present the experience via a web viewer or VR tour with annotations using A-Frame or Matterport. The good news is that low-cost and open-source solutions abound: from free phone apps to open GPU pipelines, the barrier to creating a digital twin of your stadium has never been lower. With a bit of experimentation, you can produce an interactive 3D walkthrough that lets users virtually explore the venue and learn how to use its facilities – all without breaking the bank.