Datasets:
FIORD Building_In
| Item | File | Size |
|---|---|---|
| GOFish (Ours) Gaussians | gaussians/point_cloud.ply |
489,734 Gaussians, 121 MB |
| + Geometry Objective Gaussians | gaussians_geo/point_cloud.ply |
492,218 Gaussians, 122 MB |
| + Geometry Objective mesh | mesh/mesh.ply, mesh/mesh.glb |
282 MB PLY |
| Cameras | cameras.json |
402 train and 58 test frames |
Both Gaussian files carry the 3-D Mip filter baked into their scales. Reloading each PLY reproduces the checkpoint's stored positions, rotations, opacities and spherical harmonics exactly, and its filtered scales to float32 precision. EXPORT.json holds the checks and the source checkpoint sha256.
The mesh is byte-identical to the evaluated mesh, sha256 c603b53a1a1d2b1468b1a07d272ca5931ec0e00d87eefe0b244a5d13264ba053.
lidar_alignment.json maps the mesh, Gaussians and cameras into the FIORD LiDAR ground-truth frame used for the paper's geometry scores.
See the top-level README for formats and viewers.