NeurIPS 2026 paper VODA fixes incomplete 3D Gaussian segmentation via volumetric supervision
ChongZzZhang · x · 2026-11-20
A paper accepted at NeurIPS 2026, "The Mask Is Not the Object: Volumetric Supervision for 3D Gaussian Segmentation," shows that existing 3DGS segmentation methods trained on rendered views can yield accurate 2D masks from incomplete 3D selections, since alpha-blended supervision concentrates on dominant contributors and leaves occluded Gaussians weakly trained.
Key contributions:
- VODA: a plug-and-play objective that uses local depth structure to group per-pixel contributors and directly supervises all Gaussians in selected depth clusters, bypassing contribution-weighted gradient delivery;
- Removal-3D benchmark and RDE metric for evaluating volumetric completeness via object removal;
- Across four rendering-based baselines, VODA consistently improves volumetric segmentation on Removal-3D while maintaining or improving LERF and ScanNet results; gradient reweighting and spatial propagation baselines fail to match the gains.
Related event: VODA Tackles Blind Spots in 3D Gaussian Splatting Segmentation(2 posts)→
More from Research
- PhD student seeks cheap API routing to run 20 video benchmarks, $2,000 per model — Alternative_Art2984 · 2026-11-20
- Professor builds party simulator showing why high-dimension tastes make us 'ortholonely' — prof_g · 2026-11-20
- METR Study: Developers Felt 20% Faster With AI, Actually Took 19% Longer — alifcoder · 2026-11-20
- Asuka-FM: 206M dual-stream DiT pretrained on one RTX 4090 in 160h for under $55 — bdsqlsz · 2026-11-20
- ICLR 2027 author says AC and two reviewers went silent after rebuttal experiments — Temporary_Switch_339 · 2026-11-20
- Ulms ERT training makes vision nets far easier for LLMs to edit in weight space — Timo Ropinski · 2026-11-20