Version 1 of a LEGO® bulldozer Gaussian splat, cut out of its kitchen automatically with the
same pipeline as my LEGO Bonsai Tree v2.
How it was made:
- Photos and camera positions: the "kitchen" scene of the Mip-NeRF 360 dataset (Barron et al.,
CVPR 2022).
- Segmentation: Meta's SAM 3 tracked the bulldozer through all 279 photos from the text prompts
"lego bulldozer", "yellow toy bulldozer" and "bulldozer bucket" - no manual masking. Frames
where part of the model dropped out of the mask (29 frames) were found and refilled
automatically by re-tracking from neighbouring frames, and a second pass checked every frame
for the bucket and tracks.
- Training: Nerfstudio's splatfacto-big (gsplat), 30,000 steps, trained only on the masked model.
- Clean cut-out: each Gaussian kept or removed by how much it actually paints inside vs outside
the masks across all views (the FlashSplat method). Then the underside was tidied (smeared
Gaussians under the tracks removed - the bottom was never photographed because it sat on a
table), stray clusters removed, and the model centred and scaled for viewing.
Credits:
- Mip-NeRF 360: Barron, Mildenhall, Verbin, Srinivasan & Hedman, "Mip-NeRF 360: Unbounded
Anti-Aliased Neural Radiance Fields", CVPR 2022.
- SAM 3 by Meta AI.
- Shen, Yang & Wang, "FlashSplat: 2D to 3D Gaussian Splatting Segmentation Solved Optimally",
ECCV 2024.
- 3D Gaussian Splatting: Kerbl et al., SIGGRAPH 2023. Training: Nerfstudio + gsplat. Cameras
supplied with the dataset (computed with COLMAP).
Built step by step with help from Claude (Anthropic).
LEGO® is a trademark of the LEGO Group, which does not sponsor or endorse this model.