Hi, tbh I wanted to save this for the conclusion of my upcoming third method, but I'm happy to share it briefly here as well.
My conclusion is that the Insta360 unfortunately isn't the best option for interior spaces in terms of output quality (when extracting frames from video). I'd suggest using a normal camera for the main images, and using the Insta360 more sparingly — shooting in 70MP photo mode to create sort of "checkpoints" images for coverage safety, ensuring every angle is captured.
You may have noticed that in the splat where only the Sony camera was used, the floor isn't reconstructed well (and here comes Insta360 as a solution), since I focused more on the walls and objects. I'd also suggest using a CPL filter on the lens to avoid reflections.
Additionally, in this splat I combined the raw .insv file from the Insta360 with edited JPGs from the Sony in MipMap, which isn't ideal for hybrid workflows — the system struggles to calculate common/overlapping areas between the two camera sources.
For a proper hybrid solution, I'd suggest extracting images from 8 angles of each individual 360 frame, then combining these with the main camera images to generate the point cloud (in RealityScan, for example), and training that in whichever 3DGS software you prefer. That's exactly what my third method will focus on