Research map / 3D vision / Gaussian splatting
Semantic Foam: Unifying Spatial and Semantic Scene Decomposition
Amr Sharafeldin, Aryan Mikaeili, Thomas Walker, Shrisudhan Govindarajan, Daniel Rebain, Kwang Moo Yi, Andrea Tagliasacchi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add total variation loss on Voronoi adjacency (ablation) | Semantic Foam · Average of Garden, Figurines, Horns | mIoU: 0.84 -> 0.86 | Improved |
| add total variation loss on Voronoi adjacency (ablation) | Semantic Foam · Average of Garden, Figurines, Horns | mAcc: 0.93 -> 0.94 | Improved |
| add total variation loss (ablation) | Semantic Foam · Garden (Mip-NeRF360) | mIoU: 0.86 -> 0.94 | Improved |
| add total variation loss (ablation) | Semantic Foam · Figurines (LERF-Masked) | mIoU: 0.81 -> 0.79 | Hurt |
| add total variation loss (ablation) | Semantic Foam · Horns (LLFF) | mIoU: 0.86 -> 0.86 | No clear effect |
| replace Gaussian primitives with Voronoi cells (Semantic Foam vs Gaussian Grouping) | Semantic Foam · Mip-NeRF360 (5 scenes) | mIoU: 0.75 -> 0.82 | Improved |
| replace Gaussian primitives with Voronoi cells (Semantic Foam vs SAGA) | Semantic Foam · LLFF (8 scenes) | mIoU: 0.73 -> 0.85 | Hurt |
| Semantic Foam vs Gaussian Grouping on LLFF | Semantic Foam · LLFF (8 scenes) | mIoU: 0.86 -> 0.85 | No clear effect |
| Semantic Foam vs InstanceGaussian and GARField | Semantic Foam · LERF-Masked (3 scenes) | mIoU (average): 0.61 -> 0.79 | Improved |
| add identity encoding features to Radiant Foam | Radiant Foam · Mip-NeRF360 (reconstruction) | PSNR: 29.92 -> 29.79 | No clear effect |
| add identity encoding features to Radiant Foam | Radiant Foam · LERF-Masked (3 scenes, RTX A6000) | model size (MB): 659.67 -> 782.00 | Hurt |
| add identity encoding features to Radiant Foam | Radiant Foam · LERF-Masked (3 scenes, RTX A6000) | training time (mins): 80.67 -> 81.00 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Gaussian Grouping — Gaussian Grouping: Segment and Edit Anything in 3D Scenes (ECCV 2024) · better on 4 of 6 shared results
- SAGA — SAGA: Stochastic Whole-Body Grasping with Contact (ECCV 2022) · better on 4 of 6 shared results
- GARField — GARField: Group Anything with Radiance Fields (CVPR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tackling View-Dependent Semantics in 3D Language Gaussian Splatting (ICML 2025 · similarity 0.87)
- Efficient Decoupled Feature 3D Gaussian Splatting via Hierarchical Compression (CVPR 2025 · similarity 0.87)
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.87)
- Bootstraping Clustering of Gaussians for View-consistent 3D Scene Understanding (AAAI 2025 · similarity 0.87)
- Segment then Splat: Unified 3D Open-Vocabulary Segmentation via Gaussian Splatting (NeurIPS 2025 · similarity 0.87)
- 3D Vision-Language Gaussian Splatting (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
