Research map / 3D vision / Gaussian splatting
S2GS: Streaming Semantic Gaussian Splatting for Online Scene Understanding and Reconstruction
Renhe Zhang, Yuyang Tan, Jingyu Gong, Zhizhong Zhang, Lizhuang Ma, Yuan Xie, Xin Tan
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 decoupled frozen SigLIP2 semantic stream instead of shared backbone (ablation) | S2GS · ScanNet (32 views) | mIoU: 42.17 -> 48.95 | Improved |
| use decoupled frozen SigLIP2 semantic stream instead of shared backbone (ablation) | S2GS · ScanNet (32 views) | T-SR: 38.75 -> 62.39 | Improved |
| add independent SigLIP2 semantic branch (ablation) | SIU3R · ScanNet (32 views) | mIoU: 39.98 -> 47.13 | Improved |
| add independent SigLIP2 semantic branch (ablation) | SIU3R · ScanNet (32 views) | PSNR: 17.82 -> 18.03 | No clear effect |
| add query-level contrastive learning (ablation) | S2GS · ScanNet (32 views) | T-SR: 50.13 -> 62.39 | Improved |
| add query-level contrastive learning (ablation) | S2GS · ScanNet (32 views) | mIoU: 47.13 -> 48.95 | Improved |
| add geometric distillation from causal teacher (ablation) | S2GS · ScanNet (8 views) | PSNR: 15.17 -> 20.83 | Improved |
| add geometric distillation from causal teacher (ablation) | S2GS · ScanNet (64 views) | PSNR: 10.03 -> 18.71 | Improved |
| streaming causal design vs offline global processing | S2GS vs SIU3R · ScanNet (1024 views) | per-frame time: SIU3R OOM; S2GS 0.62 s | Improved |
| pixel-wise feature lifting instead of query-based design | PFL vs S2GS · Open-vocabulary grounding | mIoU: 49.37 -> 51.83 | Mixed |
| use S2GS instead of offline methods in extremely sparse 2-view setting | S2GS · ScanNet (2 views) | PSNR: 25.79 -> 24.90 | Hurt |
| replace test-time optimization (Splat-SLAM) with feed-forward S2GS | S2GS · ScanNet (32 views) | PSNR/time: 20.88 PSNR ~345s vs 19.92 PSNR ~4s | Mixed |
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.
- SIU3R — SIU3R: Simultaneous Scene Understanding and 3D Reconstruction Beyond Feature Alignment (NeurIPS 2025) · better on 13 of 14 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 0.91)
- Efficient Decoupled Feature 3D Gaussian Splatting via Hierarchical Compression (CVPR 2025 · similarity 0.91)
- Tackling View-Dependent Semantics in 3D Language Gaussian Splatting (ICML 2025 · similarity 0.91)
- Bootstraping Clustering of Gaussians for View-consistent 3D Scene Understanding (AAAI 2025 · similarity 0.91)
- FHGS: Feature-Homogenized Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- Learning Efficient Fuse-and-Refine for Feed-Forward 3D Gaussian Splatting (NeurIPS 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.
