under the hood

Many sensors, one coherent model.

A single 360° camera gives a relative-only, scale-free splat. splathaus fuses several sensors so the result is metric, absolutely placed, and reproducible across epochs. Every device is treated as a pose and telemetry source, not just a camera.

iPhone / iPad LiDAR + ARKit motion tracking True metric scale, dense depth in tight corners, and an excellent 6-DoF trajectory, the foundation of every capture.
360° camera Dual-fisheye + stabilisation gyro Full coverage of a space in one pass, plus a camera-motion prior. An Insta360 X5 is the supported rig.
Drone Gimbal + flight log Aerial coverage of the roofline with pose priors for the merged set. A DJI Mavic 4 Pro or comparable airframe.
Ground markers Surveyed once Centimetre absolute anchors every epoch registers to.

The fusion stack.

No single source is enough; the fusion is the point. Some layers are live today; others are on the roadmap, marked here so you always know what’s shipping.

Unified telemetry

Every device emits a timestamped pose / IMU / GNSS / gimbal stream into one session, the foundation the rest of the fusion hangs off.

LiDAR metric scale

In progress

iPhone LiDAR rescales the 360° splat to true-world dimensions; surveyed markers fix the absolute anchor.

Raw-GNSS PPK

Planned

On supported Android devices, raw dual-band GNSS post-processed against Japan's national CORS network gives decimetre geotags that tie epochs together, a precision option beyond the LiDAR-and-marker baseline.

Pose-prior SfM

Planned

Feed VIO, GNSS and gimbal poses into COLMAP / GLOMAP as priors so air↔ground registration converges where image matching alone fails.

Depth-supervised 3DGS

Planned

Use iPhone LiDAR depth as training supervision for fewer floaters and cleaner geometry in sparse aerial coverage.

Face & plate blur

Planned

Automatic redaction of faces and licence plates before anything exterior is ever published, for PIPA compliance.

Metric accuracy

A 360° pass alone is scale-free. iPhone LiDAR rescales it to true-world dimensions; a pair of coded ground markers fixes absolute scale; extra markers act as accuracy checkpoints. LiDAR and the markers are the metric backbone. Where extra absolute precision is wanted, raw-GNSS post-processing (on supported Android devices) adds a decimetre-level global tie across epochs: a coarse prior, not survey-grade, with the markers remaining the centimetre anchor.

Privacy by structure

Every capture lands in one of two buckets, public or private, at ingest. Exterior and street captures can become shareable splats; interior footage is structurally separated and never published. The boundary is enforced in code, and a deploy-time gate refuses to ship any asset that isn’t on a reviewed public allowlist.

Delivered everywhere.

Web

.splat / .spz stream straight into the browser, the viewer on this site is exactly that.

VR

.spz / .ply for standalone and tethered VR headsets, walk the property at 1:1 scale.

Cinematic

A camera-path render handed to Blender / nerfstudio for a polished fly-through.