360 Splat Pro v1.4.0 Adds 360 Plus Perspective Camera Fusion for Gaussian Splatting
Michael Rubloff
Ronski has released 360 Splat Pro v1.4.0, and now it is possible to combine a single 360 capture with ordinary perspective cameras, phones, DSLRs, drones, into one reconstruction. The 360 footage supplies full spherical coverage while the perspective cameras drop sharp, high resolution detail wherever it is wanted, and the tool now ships two distinct workflows for getting there, chosen entirely by the format the 360 footage arrives in.
This builds directly on the COLMAP based 360 to 3DGS preparation pipeline the tool launched with and extended through RealityScan meshing in v1.2 and GPU friendly alignment plus raw Insta360 photo support in v1.2.5.
Raw Fisheye Fusion takes raw dual fisheye footage straight off the camera. It solves the raw fisheye frames on their own first, then projects each frame into five cube faces, declares those faces a locked rig, and lets every perspective camera join as its own free camera, using exhaustive matching. It is a two stage process. Reconstruct the 360, then fuse the ground cameras. Equirect Fusion handles equirectangular (2:1) footage. It renders each panorama into twelve perspective views and reconstructs everything in a single pass, with the twelve views forming the rig and the perspective cameras again joining as free cameras, using sequential plus vocabulary matching.
The technical reason both workflows refuse to match a fisheye frame directly against a phone photo is the geometry is too different and feature matching falls apart, so each 360 frame's views are declared a zero-baseline rig and each ground camera becomes a free camera, with a single rig constrained solve tying everything together. Ronski reports the latest test fused six cameras at once at 100 percent registration.
The rest of v1.4.0 is workflow polish. The release adds notifications via ntfy (sound, mobile, or email, configurable per single run, batch, or error state), a Skip AI Masking checkbox for cases where only shape or paint masks are needed and a fix for multi camera RealityScan import, which previously let only the first of several videos through and now imports and solves them all.
This is the plugin I use most often for my 360 work and I'm excited to see its development continued. The plugin can be purchased here.
