Research map / 3D vision / Gaussian splatting
AG2aussian: Anchor-Graph Structured Gaussian Splatting for Instance-Level 3D Scene Understanding and Editing
Zhaonan Wang, Manyi Li, Changhe Tu
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 anchor-graph propagation loss and graph-based region growing together (ablation) | AG2aussian · LERF-OVS text query | mIoU: 42.72 -> 54.35 | Improved |
| add graph propagation loss L_prop (ablation) | AG2aussian · LERF-OVS text query | mIoU: 42.72 -> 49.10 | Improved |
| add graph-based region growing (GraphSeg) (ablation) | AG2aussian · LERF-OVS text query | mIoU: 42.72 -> 45.85 | Improved |
| add L_prop on top of GraphSeg (ablation) | AG2aussian · LERF-OVS text query | mIoU: 45.85 -> 54.35 | Improved |
| constrain Gaussian position and scale within anchor voxel (localization) (ablation) | AG2aussian · LERF-OVS text query | mIoU: 30.45 -> 54.35 | Improved |
| use anchor structure instead of learnable codebook (Super-Gaussian style) (ablation) | AG2aussian · LERF-OVS text query | mIoU: 40.51 -> 54.35 | Improved |
| use anchor-graph structured Gaussians instead of free Gaussians | OpenGaussian-style contrastive training · LERF-OVS | memory (GB): 16.81 GB -> 7.56 GB | Improved |
| use AG2aussian instead of SAGA for kitchen click query | AG2aussian · LERF-OVS click query kitchen | mIoU: 83.03 -> 80.80 | Hurt |
| use AG2aussian instead of GsGrouping for text query on Mip-NeRF360 bonsai | AG2aussian · Mip-NeRF360 bonsai text query | mIoU: 68.73 -> 53.47 | Hurt |
| use AG2aussian instead of SAGA in training time | AG2aussian · LERF-OVS training time | training time (min): 33.63 -> 39.55 | Hurt |
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.
- SAGA — SAGA: Stochastic Whole-Body Grasping with Contact (ECCV 2022) · better on 6 of 8 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
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 2025 · similarity 0.92)
- FHGS: Feature-Homogenized Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
- GaussianCut: Interactive segmentation via graph cut for 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.91)
- Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields (CVPR 2024 · similarity 0.90)
- Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration (CVPR 2025 · similarity 0.90)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 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.
