pretty_name: SplatCast sample scenes
license: other
license_name: cc-by-4.0-and-mit
license_link: LICENSE
tags:
- gaussian-splatting
- 3d-gaussian-splatting
size_categories:
- n<1K
SplatCast sample scenes
The trained 3D Gaussian Splatting scenes that SplatCast, a Mac app that trains splats on Apple silicon, shows in its Inspiration gallery. The app downloads a scene from here the first time someone opens it.
Each folder holds scene.splatcast, its credits (ATTRIBUTION.md), its licence and scene-provenance.json.
Format. .splatcast is SplatCast's compact scene format: a 24-byte header ("SCST", version, splat count, SH coefficient count, flags, SH scale) followed by an LZFSE-compressed block of float32 positions, float16 scales, rotations, opacities and DC colour, and int8 higher-order spherical harmonics. On Saint Joseph's Oratory it is 4× smaller than the PLY and renders at 67 dB from the original.
| Folder | Scene | Splats | Licence |
|---|---|---|---|
ClockTower |
Clock Tower, Montreal | 1,150,558 | CC BY 4.0 |
Oratory |
Saint Joseph's Oratory, Montreal | 1,261,862 | CC BY 4.0 |
CityHall |
Montreal City Hall | 1,928,048 | CC BY 4.0 |
Riverview |
Riverview Room (Eyeful Tower) | 3,274,372 | MIT |
EyefulApartment |
Apartment (Eyeful Tower) | 3,169,091 | MIT |
Credits. The Montreal scenes were trained from photos in Heritage3DMtl by Rucha Shende, Sudhir Mudur, Tiberiu Popa and Kaustubha Mendhurwar, Concordia University (https://zenodo.org/records/17063337), CC BY 4.0; the trained scenes are offered under the same licence. The Eyeful Tower scenes: copyright (c) Meta Platforms, Inc. and affiliates (https://github.com/facebookresearch/EyefulTower), MIT licence. Changes are described in each folder's ATTRIBUTION.md. No endorsement by the photographers, Concordia University, Meta or the places shown is implied.