Four assets. One frame. Open formats.

A mesh and a point cloud from every input, and on request a Gaussian splat and a navigable 360° tour. The mesh arrives as raw and textured GLB, the splat as .ply and .spz, the point cloud as a binary PCD at 5 cm spacing, and the tour as equirectangular panoramas carrying position, orientation and neighbour links.

The point cloud is the one asset that does not share the mesh's frame. map.pcd is right-handed with +Z up, while the mesh and the pose files are left-handed with +Y up. Swap the Y and Z components of every point, so (x, y, z) becomes (x, z, y). The swap flips handedness, which is what makes the two agree. A rotation on its own leaves a mirrored cloud. One conversion works across every map in your account.

Because the file being rendered is the one you uploaded. The splat MultiSet returns is transformed into the map's coordinate frame so it sits exactly on the generated mesh. Your original sits in whatever arbitrary frame your reconstruction tool used. Render the returned .ply or .spz and everything registers.

No. Accuracy is computed from the map's feature data, not from the visual assets. Clean up or re-texture a mesh in Blender and push it back with a mesh upload at any time, without re-processing the map. Keep the origin at (0, 0, 0) or everything placed against it will be offset once poses arrive.

360 video. It is the only input that can produce all four assets from one capture: a mesh, a point cloud, a splat and a tour. E57 and MatterPak scans carry their own panoramic imagery, so they can produce a tour but no splat. LiDAR captures from the MultiSet App and Unity SDK have no panoramas, so no tour is available for them at any setting.

Yes. Everything the Developer Portal viewer does is available over REST: list the maps that have tours, load a complete tour in one call, or fetch just the neighbourhood around a user as they move. The files are standard formats, so nothing is tied to our renderer.