Here we have the result of a pretty boiler plate example of open-source gaussian splat scene generation. This was rendered via RealityScan for the sparse point cloud mapping and image extraction, then onto LichtFeld Studio for splat generation. I use a combination of a thread ripper cpu and an RTX 5070ti so this only took around 3 minutes to compute.
I wanted this experiment to utilise datasets that are not of pristine origin. So many spats source from existing point cloud datasets and newer high resolution cameras rigs or drones. The software required to generate these have become much more flexible in what can be ingested and I was curious if incomplete or unreal datasets could make something interesting.