Research map / 3D vision / Gaussian splatting
LUDVIG: Learning-Free Uplifting of 2D Visual Features to Gaussian Splatting Scenes
Juliette Marrie, Romain Menegaux, Michael Arbel, Diane Larlus, Julien Mairal
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 |
|---|---|---|---|
| uplift features into 3D and render instead of single-view 2D prediction (ablation) | DINOv2/SAM2 segmentation · SPIn-NeRF multi-view segmentation | IoU: 88.5 -> 91.6 (DINOv2); 88.6 -> 93.8 (SAM2) | Improved |
| add 3D graph diffusion to uplifted DINOv2 features (ablation) | DINOv2 segmentation · SPIn-NeRF multi-view segmentation | IoU: 91.6 -> 93.8 | Improved |
| add SAM segmentation to CLIP relevancy with graph diffusion (ablation) | LUDVIG CLIP segmentation · LERF object segmentation | IoU overall: 50.4 -> 64.3 | Improved |
| add graph diffusion to raw CLIP relevancy (ablation) | LUDVIG CLIP segmentation · LERF object segmentation | IoU overall: 33.6 -> 50.4 | Improved |
| add graph diffusion on top of SAM-based segmentation (ablation) | LUDVIG CLIP segmentation · LERF object segmentation | IoU overall: 57.3 -> 64.3 | Improved |
| run more iterations of reconstruction-loss optimization instead of single aggregation step (ablation) | Feature uplifting · feature quality | feature quality | No clear effect |
| filter out half of the Gaussians by importance (ablation) | LUDVIG · all tasks | quality | No clear effect |
| use learning-free uplifting instead of LangSplat feature learning | Open-vocabulary segmentation · LERF 2D object segmentation | runtime (min): 105 -> 10 | Improved |
| replace OpenGaussian feature training with direct uplifting | OpenGaussian pipeline · ScanNet semantic segmentation (19 classes) | mIoU: 24.7 -> 33.9 | Improved |
| use LUDVIG with DINOv2 and SAM masks instead of SAGA/OmniSeg3D | Multi-view segmentation · SPIn-NeRF multi-view segmentation | IoU: 94.3 -> 93.8 (vs OmniSeg3D) | No clear effect |
| use purely geometric mask reprojection on 360-degree scenes (ablation) | Multi-view segmentation baseline · SPIn-NeRF multi-view segmentation | IoU: 93.8 -> 73.1 | Hurt |
Later papers that compared with this method
3 later accepted papers report LUDVIG as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- LightSplat: Fast and Memory-Efficient Open-Vocabulary 3D Scene Understanding in Five Seconds (CVPR 2026 · better on 6 of 6 shared results)
- ZeroSplat: Generalized Referring Segmentation in 3D Gaussian Splatting (ECCV 2026 · better on 3 of 3 shared results)
- ExtrinSplat: Decoupling Geometry and Semantics for Open-Vocabulary Understanding in 3D Gaussian Splatting (CVPR 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 4 on most shared results.
- LangSplat — LangSplat: 3D Language Gaussian Splatting (CVPR 2024) · better on 3 of 3 shared results
- OmniSeg3D — OmniSeg3D: Omniversal 3D Segmentation via Hierarchical Contrastive Learning (CVPR 2024) · better on 1 of 2 shared results
- LERF — LERF: Language Embedded Radiance Fields (ICCV 2023) · better on 2 of 2 shared results
- SAGA — SAGA: Stochastic Whole-Body Grasping with Contact (ECCV 2022) · better on 1 of 2 shared results
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
- Efficient Decoupled Feature 3D Gaussian Splatting via Hierarchical Compression (CVPR 2025 · similarity 0.89)
- Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields (CVPR 2024 · similarity 0.89)
- FHGS: Feature-Homogenized Gaussian Splatting (NeurIPS 2025 · similarity 0.89)
- Bootstraping Clustering of Gaussians for View-consistent 3D Scene Understanding (AAAI 2025 · similarity 0.88)
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 2025 · similarity 0.88)
- GaussianCut: Interactive segmentation via graph cut for 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.88)
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.
