Comparison of 3D Models made from 35s of Video by RealityCapture, LumaAI, and PostShot

Quick comparison of 3 different software to create a 3d model from video. Using just a 35s clip from an FPV flight with a DJI O3 camera. Surprising what you can get! RealityCapture has the least detail but also the least input to be fair, and also the quickest to create. Final model output will cost something, but I believe one would have the most options.
For zero effort LumaLabs AI did a wonderful job. Had to wait about a day for the processing to happen in their cloud, but it was also free. I'm sure that will change in the future. I remember the early days of DroneDeploy.
PostShot looks the best and most realistic in this scenario. Processing time was about twice RealityCapture's. Free software (beta) with a good user guide, I basically went with their recommended settings but used full res images and 50k steps. Has a keyframe video timeline function to create fly-throughs.
With more deliberate planning better models could have easily been created out of all three. I don't think there's a clear winner, as much depends on your output needs. RealityCapture felt very easy for me as I have some experience with it and the software has been around longer, with plenty of options on tap an YT vids. The quality coming out of PostShot makes me want to revisit it with better footage.
#3dmodel #radiancefields #nerf #gaussiansplat #photogrammetry #lumaai
#postshot #realitycapture #drone