Splat Generator
The viewer is a 3D stage with a floating menu — it wants a wide screen, and a scene gets in by dropping a splat file onto it, which is easiest in its own window on a desktop.
Open anywayNothing you open is uploaded. There is no upload path in the tool at all: scenes are read in the tab and never leave the machine.
DL-SplatGenerator reads Gaussian-splat scenes of landscapes and sites — and measures them: distances, heights, areas and sections taken from a photographed place, not a drawn one.
A reconstruction, not a survey.
A measuring splat viewer · open source (Apache‑2.0) · v1 · runs in the browser, nothing to install
A splat scene is built from photographs, and it is convincing enough to forget that. This viewer is arranged so you cannot: the scene carries no scale of its own until you set one against a distance you know, and every measurement after that names the reconstruction it was read from. What it offers is the useful half of a site visit you can repeat at the desk — standing in the place, taking dimensions off it, leaving notes where they belong.
A scene gets in by dropping a .ply, .spz or .splat file onto the stage — from any Gaussian-splatting pipeline. The same open-source project includes the generator that makes such scenes from site video, run locally on your own machine; what you are using here is the viewer.
One scene, several readings
Measuring is the point, so the instruments are the feature list. Each one reads the same reconstruction a different way, and everything stays in the scene where you put it.
Distances
Point to point through the scene — the tape measure, after the scale is set.
Heights
Vertical rise against the scene’s up axis — a wall, a bank, a terrace step.
Areas
A surface traced point by point and read as one figure.
Sections
A box cuts the scene open, and the ground shows its profile.
Plan view
The scene from straight above — the drawing convention, recovered from photographs.
Notes
Observations pinned in space, where they were made rather than in a list.
Viewpoints & stills
Named positions to return to, and framed images taken from them.
Take it into Blender
A solved capture does not have to end in the viewer. Capture Walk, the companion Blender add-on, imports a solved capture with the walk that was filmed — the splat scene together with the camera path as a real animated camera — so the reconstruction can be modelled against, lit and rendered inside Blender.
Download Capture Walk v0.21.0 — one file for Windows and macOS (Blender 4.5 LTS or newer; 4.5 is the last Blender for Intel Macs). Install in Blender via Edit → Preferences → Get Extensions → Install from Disk. Licensed GPL‑2.0‑or‑later — an add-on is a derived work of Blender — with its licence and credits inside the file, which is also the complete source.
How it is checked
The viewer ships with its own acceptance page — 28 checks that run in the browser, against the deployed tool itself. It loads a calibration scene whose marker spans are known by construction, picks every marker through the renderer, sets the scale on a span that is exactly one metre, and then measures distances and areas against their true lengths — and it puts the section box through the failure it once had. All 28 pass on the version running above.
Checks are not the same as a surveyor’s scepticism. Verify against a known dimension before you build on a measurement.
Take it apart
| Source | github.com/DLandscapes/dl-splat-generator |
|---|---|
| Version | v1, untagged — a DOI follows once the metadata has settled. |
| Licence | Apache‑2.0. The libraries it uses are MIT-licensed and the typefaces keep their own OFL licences, travelling beside the files they cover. |
| Data | None is shipped and none is collected. Scenes you open are read in the tab and never leave your machine. |
Provided as is, without warranty of any kind — see the Apache‑2.0 licence for the full terms. A splat scene is a reconstruction from photographs: set the scale against a known distance, and verify any measurement you intend to act on against the site itself.
Developed by Digital Landscapes with AI assistance. The design decisions and all verification are Digital Landscapes’ own.