Tabletop clutter scene
scene_tabletop · Procedural (exact GT, offline) ·
generated locally with exact ground truth
Ground-truth mesh
Drag to orbit, scroll to zoom. Rendered from gt/mesh.obj
(4788 faces).
Ground-truth views
lighting:
channel:
Why this sample is in the benchmark
Several objects with mutual occlusion on a support surface. This is the first sample where the geometry of one object is inferred from observations that are partly missing, and where "floater" artifacts between objects become visible.
Ground truth
- GT type
- exact
- licence
- Generated locally -- no third-party licence
- source
- Procedural (exact GT, offline)
- builder
- tabletop_scene
- faces
- 4788
- vertices
- 2474
- watertight
- False
- surface area
- 18.4279 m²
- extents (m)
- [2.4, 1.135, 2.4]
- GT points
- 20000 surface samples
- views
- 16 × 128px (orbit, fov 50°)
- lighting
- studio, warm_key, low_key
Acquisition plan
kind local, automatic True,
estimated size 4.32 MiB
- Build geometry 'tabletop_scene' with {}
- Render 16 views (orbit @ 128px, fov 50 deg)
- Shade under lighting: studio, warm_key, low_key
- Write GT mesh, surface points, depth, normals, poses
Fully offline. Ground truth is exact (no SfM, no scan noise).
Comparisons
| method | F-score | Chamfer-L1 | scale | |
|---|---|---|---|---|
mock-3dgs |
0.3700 | 0.00707 | 0.9258 | open |