MultiSet ingests structured E57 files with embedded panoramic images from the leading reality-capture hardware.
Upload your existing scans. No format conversion. No proprietary middleware. No rescanning.
Upload the E57 and process it. There is no conversion step to run first, no proprietary re-capture, and no requirement that the scan was taken with AR in mind. E57 is the industry interchange format, which is exactly why it is the right input: whatever surveyed the building almost certainly exported one.
What comes back is two different things from the same capture. A machine-readable map, which is the feature representation a camera localizes against, and a set of human-readable artifacts, which are the mesh, splat, panorama tour and point cloud people actually look at. Most of the market produces one or the other. The table below is the full list.
| Input | Source hardware or tool | Produces | Notes |
|---|---|---|---|
| E57 | Matterport, Leica, NavVis, Faro, XGRIDS | VPS map, point cloud | The industry interchange format. Most existing scans are already here |
| MatterPak | Matterport | VPS map, mesh | The bundle Matterport exports for downstream use |
| GLB / PLY | Polycam, Scaniverse, CAD | VPS map, mesh | Textured meshes |
| Gaussian splat (.ply / .spz) | Any 3DGS pipeline | VPS map, splat | Must be metric-scaled. A scale-free splat is a picture, not a map |
| 360 video | Insta360 and similar, raw or equirectangular MP4 | VPS map, panorama tour | One walkthrough, or one flight |
| Native LiDAR | iPhone and iPad Pro, via MultiSet Mapper | VPS map, mesh | No professional hardware required |
Outputs from a single capture: a textured mesh (.glb), a Gaussian splat (.ply or .spz), a 360 panorama tour, and a point cloud (.pcd at 5 cm spacing), plus the VPS feature map itself, which stays inside the platform.
Scanner grade matters far less than most buyers expect. What decides whether a map localizes reliably is how the space was walked, and four things account for nearly all of it.
Coverage overlap. Consecutive viewpoints need to share enough of the same geometry for the reconstruction to tie together. Gaps between stations are the most common cause of a map that works in some places and fails in others, and they are invisible in a rendered preview.
Lighting consistency. A capture taken across a whole day, or half in daylight and half under sodium lamps, produces a map that disagrees with itself. Capture a space in one lighting condition where you can, and expect a repeat pass for spaces that change dramatically between shifts.
Texture. Localization needs something to recognise. Long blank corridors, uniform white walls, freshly painted plant rooms and empty aisles are the hard cases, not the large ones. A cluttered 10,000 sq ft warehouse is easier than a bare 2,000 sq ft corridor.
Capture height. Map at roughly the height the device will be used at. A tripod scan at 1.5 m serves a phone or a headset well and serves a floor-level AMR poorly, because the two are looking at different parts of the room.
Scanner-specific guides: Matterport, NavVis, Leica, FARO and XGRIDS.
None of these require better hardware. They require the capture to be planned for how the map will be queried, which is a briefing question rather than a budget one.
Where this lands in practice: oil and gas builds one plant frame from tank farm to unit interior out of the E57 an as-built survey already produced, using the right instrument per zone and stitching the result into a single frame.
The same holds before a building is finished: construction, where the as-built E57 the project already paid for becomes the frame field users localize against, and a partial re-scan merges in as a version rather than a separate map.
