Blender synthetic -- Ficus
blender_ficus · Blender / NeRF synthetic ·
planned — not present locally yet
Ground-truth mesh
Not present locally. See the acquisition plan below.
Why this sample is in the benchmark
Hundreds of thin leaves: the hardest geometry in the classic object set and the one that most cleanly separates a method that recovers thin structure from one that produces fuzzy volumes.
Ground truth
- GT type
- exact
- licence
- MIT / research (see release)
- source
- Blender / NeRF synthetic
- builder
- - {'scene': 'ficus'}
- views
- 200 × 800px (orbit, fov 50°)
- lighting
- provided
Acquisition plan
kind script, automatic False,
estimated size 80 MiB
- Auto-download through nerfbaselines (recommended)
pip install nerfbaselines && python -c "import nerfbaselines; nerfbaselines.download_dataset('blender')" - Original NeRF synthetic + LLFF data release
- Export ground-truth mesh from the .blend file
blender -b ficus.blend --python examples/export_blend_mesh.py
Blender / NeRF synthetic scenes (Mildenhall et al.): lego, chair, drums, ficus, hotdog, materials, mic, ship. 100-200 upward-facing views with exact poses (transforms.json) AND the .blend files, from which the ground-truth mesh can be exported exactly. This is the reference object-level geometry benchmark.
Comparisons
No reconstruction compared yet. Run
geombench compare blender_ficus --mock or point it at a real
3DGS output.