Research map / 3D vision / Gaussian splatting
SplatTalk: 3D VQA with Gaussian Splatting
Anh Thai, Songyou Peng, Kyle Genova, Leonidas Guibas, Thomas Funkhouser
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 |
|---|---|---|---|
| use entropy-adaptive Gaussian sampling instead of random sampling (ablation) | SplatTalk · ScanQA (val) | Avg: 26.73 -> 26.90 | Improved |
| use entropy-adaptive sampling instead of point-density sampling (ablation) | SplatTalk · ScanQA (val) | Avg: 26.00 -> 26.90 | Improved |
| use entropy-adaptive sampling instead of FPS (ablation) | SplatTalk · ScanQA (val) | Avg: 26.78 -> 26.90 | Improved |
| use entropy-adaptive sampling instead of FPS (ablation) | SplatTalk · ScanQA (val) | EM@1: 17.7 -> 17.1 | Hurt |
| increase visual input from 729 to 32,076 tokens (ablation) | SplatTalk · ScanQA (val) | EM@1: 16.2 -> 17.1 | Improved |
| increase visual input from 729 to 32,076 tokens (ablation) | SplatTalk · SQA3D (test) | EM@1-R: 38.0 -> 41.3 | Improved |
| increase visual input from 729 to 32,076 tokens (ablation) | SplatTalk · MSR3D (test) | EM@1: 7.9 -> 14.1 | Improved |
| use 3D Gaussian features instead of 2D LLaVA-OV features | SplatTalk · ScanQA (val) | CIDEr: 50.0 -> 61.7 | Improved |
| fine-tune SplatTalk on ScanQA | SplatTalk · ScanQA (val) | CIDEr: 61.7 -> 77.1 | Improved |
| fine-tune SplatTalk on ScanQA and SQA3D | SplatTalk · SQA3D (test) | EM@1: 26.1 -> 47.6 | Improved |
| fine-tune SplatTalk on ScanQA | SplatTalk · MSR3D zero-shot | Overall LLM-Match: 41.8 -> 41.5 | No clear effect |
| fine-tune SplatTalk on ScanQA and SQA3D | SplatTalk · MSR3D zero-shot | Overall LLM-Match: 41.8 -> 26.5 | Hurt |
Later papers that compared with this method
2 later accepted papers report SplatTalk as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- LLaVA³: Representing 3D Scenes Like a Cubist Painter to Boost 3D Scene Understanding of VLMs (AAAI 2026 · better on 3 of 3 shared results)
- GenSplat: Bridging the Generalization Gap in 3DGS Language Comprehension (CVPR 2026 · 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
- 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 0.91)
- Chain of Semantics Programming in 3D Gaussian Splatting Representation for 3D Vision Grounding (CVPR 2025 · similarity 0.90)
- LangSplat: 3D Language Gaussian Splatting (CVPR 2024 · similarity 0.89)
- 3D Vision-Language Gaussian Splatting (ICLR 2025 · similarity 0.89)
- 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.
