Research map / 3D vision / Gaussian splatting
LESV:Language Embedded Sparse Voxel Fusion for Open-Vocabulary 3D Scene Understanding
Fusang WANG, Nathan Piasco, Moussab Bennehar, Luis G Roldao Jimenez, Dzmitry Tsishkou, Fabien Moutarde
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 monocular patch-depth and normal supervision to SVRaster (ablation) | LESV · Architecture ablation (SAM+CLIP features), ScanNet 19 classes | mIoU (%): 30.62 -> 31.37 | Improved |
| Add multi-level TSDF depth confidence gating (ablation) | LESV · Architecture ablation (SAM+CLIP features), ScanNet 19 classes | mIoU (%): 31.37 -> 32.06 | Improved |
| Add depth confidence on LERF retrieval (ablation) | LESV · Architecture ablation, LERF 3D retrieval | mIoU (%): 53.32 -> 53.37 | No clear effect |
| Swap SAM+CLIP for AM-RADIO in Dr. Splat (ablation) | Dr. Splat · Backbone ablation, ScanNet 19 classes | mIoU (%): 31.66 -> 37.57 | Mixed |
| Use dense AM-RADIO patch features in LESV vs SAM+CLIP features (ablation) | LESV · Backbone ablation, ScanNet 19 classes | mIoU (%): 32.06 -> 53.22 | Improved |
| Replace deep upsampler (AnyUp) with sliding window + SCRA/SCGA (ablation) | LESV · Upsampling ablation, LERF 3D retrieval | mIoU (%): 48.02 -> 56.11 | Improved |
| Add SCRA and SCGA self-correlating aggregation (ablation) | LESV · Upsampling ablation, LERF 3D retrieval | mIoU (%): 49.79 -> 56.11 | Improved |
| Sliding window upsampling without SCRA/SCGA vs raw features (ablation) | LESV · Upsampling ablation, LERF 3D retrieval | mIoU (%): 51.19 -> 49.79 | Hurt |
| SVRaster-based LESV vs Dr. Splat on ScanNet-19 | LESV · ScanNet 19 classes | mIoU (%): 31.66 -> 53.22 | Improved |
| LESV vs 2D-specialised methods on LERF 2D retrieval | LESV · LERF 2D object retrieval (4 scenes avg) | mIoU (%): 61.28 -> 56.84 | Hurt |
| Dense feature lifting vs hierarchical multi-level masks | LESV · LERF 3D object retrieval (4 scenes avg) | mIoU (%): 58.82 -> 56.11 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- LangSplat — LangSplat: 3D Language Gaussian Splatting (CVPR 2024) · better on 2 of 2 shared results
- GAGS — GAGS: Granularity-Aware Feature Distillation for Language Gaussian Splatting (AAAI 2026) · better on 1 of 1 shared result
- LangSplatV2 — LangSplatV2: High-dimensional 3D Language Gaussian Splatting with 450+ FPS (NeurIPS 2025) · better on 0 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
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.91)
- Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration (CVPR 2025 · similarity 0.90)
- Language Embedded 3D Gaussians for Open-Vocabulary Scene Understanding (CVPR 2024 · similarity 0.90)
- SOVGaussian: Sparse-View 3D Gaussian Splatting for Open-Vocabulary Scene Understanding (AAAI 2025 · similarity 0.90)
- 3D Vision-Language Gaussian Splatting (ICLR 2025 · similarity 0.90)
- Tackling View-Dependent Semantics in 3D Language Gaussian Splatting (ICML 2025 · similarity 0.89)
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.
