Research map / 3D vision / Gaussian splatting
ExtrinSplat: Decoupling Geometry and Semantics for Open-Vocabulary Understanding in 3D Gaussian Splatting
Jiayu Ding, Xinpeng Liu, Zhiyi Pan, Shiqiang Long, Ge Li
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 entropy filter and opacity filter together for neutral points (ablation) | ExtrinSplat · LERF mean | mIoU (%): 53.0 -> 54.3 | Improved |
| add entropy filter alone (ablation) | ExtrinSplat · LERF mean | mIoU (%): 53.0 -> 53.2 | No clear effect |
| add opacity filter alone (ablation) | ExtrinSplat · LERF mean | mIoU (%): 53.0 -> 52.6 | Hurt |
| use VLM-generated text hypotheses instead of single-view CLIP image feature (ablation) | ExtrinSplat · LERF mean | mIoU (%): 36.9 -> 54.3 | Improved |
| use VLM text hypotheses instead of averaged all-view CLIP image features (ablation) | ExtrinSplat · LERF mean | mIoU (%): 39.2 -> 54.3 | Improved |
| use VLM text hypotheses instead of IoU-filtered averaged CLIP image features (ablation) | ExtrinSplat · LERF mean | mIoU (%): 50.1 -> 54.3 | Improved |
| use stronger VLM (Gemini 2.5 Pro vs InternVL3-78B vs SenseNova 6.5 Pro) (ablation) | ExtrinSplat · LERF mean | mIoU (%): 47.0 -> 50.2 -> 54.3 | Improved |
| increase number of text hypotheses per instance from 1 to 5 (ablation) | ExtrinSplat · LERF mean | mIoU (%): 44.0 -> 54.3 | Improved |
| increase number of text hypotheses from 5 to 10 (ablation) | ExtrinSplat · LERF mean | mIoU (%): 54.3 -> 53.6 | Hurt |
| set opacity threshold to 0 (remove all high-entropy points) (ablation) | ExtrinSplat · LERF mean | mIoU (%): 54.3 -> 53.2 | Hurt |
| use extrinsic index layer instead of embedding per-Gaussian features | ExtrinSplat · LERF figurines | CLIP feature storage and VRAM: ~3GB -> ~3MB; ~8 GB peak VRAM | Improved |
| use ExtrinSplat vs best prior training-free method on LERF | ExtrinSplat · LERF mean | mIoU (%): 50.4 -> 54.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LUDVIG — LUDVIG: Learning-Free Uplifting of 2D Visual Features to Gaussian Splatting Scenes (ICCV 2025) · 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 0.91)
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
- Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration (CVPR 2025 · similarity 0.90)
- Bootstraping Clustering of Gaussians for View-consistent 3D Scene Understanding (AAAI 2025 · similarity 0.90)
- ReferSplat: Referring Segmentation in 3D Gaussian Splatting (ICML 2025 · similarity 0.90)
- Segment then Splat: Unified 3D Open-Vocabulary Segmentation via Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
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.
