A Gaussian splat of the LEGO® Bonsai Tree (set 10281), cut out of its room automatically.
How it was made:
- Photos and camera positions: the "bonsai" scene of the Mip-NeRF 360 dataset (Barron et al.,
CVPR 2022), used with the permission of the authors.
- Segmentation: Meta's SAM 3 tracked the tree, pot and wooden stand through all 292 photos from
text prompts ("bonsai tree", "plant pot", "wooden stand"), with the cloth excluded. Gaps where
the tree or pot dropped out were refilled automatically by re-tracking from neighbouring frames.
- Training: Nerfstudio's splatfacto-big (gsplat) trained only on the masked object, 30,000 steps.
- Clean-up: each Gaussian was labelled object/background by how much it actually paints inside
vs outside the masks across every view - the method from "FlashSplat: 2D to 3D Gaussian
Splatting Segmentation Solved Optimally" (Shen, Yang & Wang, ECCV 2024) - then centred.
Built with help from Claude (Anthropic).
LEGO® is a trademark of the LEGO Group, which does not sponsor or endorse this model.