Gaussian Splat Test: Around Salem, MA

12 April 2024 Edit: The following info about Gaussian Splats is inaccurate. It’s my first raw assessment capturing Splats and using the Kiri engine. I’m leaving it up for provenance and possibly well deserved ridicule. For instance: as with photogrammetry the more quality data or captured content you have the better. Quality being the key word here. If  your images are all over the place with the camera settings you will get poor results. I had my camera on auto at this time. That’s what gave the bad results with the F-stop and light metering changing with the extra orbits. Always lock your camera settings when capturing an area.

I have been exploring the use of Gaussian Splats over the holiday and have become addicted to capturing whatever I can with some interesting first results. Notice the reflection of the binocular stand in the puddle, this would be impossible to capture with photogrammetry & LiDAR. Unlike photogrammetry it looks like better results are from less data meaning instead of many orbits around an object or location at different points just one good one is best with Gaussian Splats. More than one orbit tends to muddy the subject. I’m guessing that is something that will be fixed in the future. https://www.kiriengine.app/ running on a M1 iPad pro so far has been the easiest app to scan/capture content for me although I notice I get better results if I use the iPad default camera software for taking photos or videos and than uploading those to Kiri engine.