Research map / 3D vision / Gaussian splatting
S3R-GS: Streamlining the Pipeline for Large-Scale Street Scene Reconstruction
Guangting Zheng, Jiajun Deng, Xiaomeng Chu, Yu Yuan, Houqiang Li, Yanyong Zhang
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 NeuralODE motion model to 2D-box decomposition (ablation) | S3R-GS · KITTI NVS benchmark (75%) | PSNR: 21.09 -> 30.48 | Improved |
| replace 3D boxes with 2D boxes + NeuralODE (ablation) | S3R-GS · KITTI NVS benchmark (75%) | PSNR: 31.49 -> 30.48 | Hurt |
| use naive 2D boxes with learnable poses instead of 3D boxes (ablation) | S3R-GS · KITTI NVS benchmark (75%) | PSNR: 31.49 -> 21.09 | Hurt |
| add BEV-semantic initialization augmentation (ablation) | S3R-GS · KITTI 0011 (75%) | PSNR: 24.21 -> 24.66 | Improved |
| add BEV-semantic initialization augmentation (ablation) | S3R-GS · KITTI NVS benchmark (75%) | PSNR: 31.49 -> 31.54 | Improved |
| add BEV-semantic initialization augmentation (ablation) | S3R-GS · KITTI 0011 (75%) | LPIPS: 0.184 -> 0.171 | Improved |
| use S3R-GS streamlined pipeline instead of 4DGF pipeline | S3R-GS · Argoverse 2 | reconstruction time (hours): 54.39 h -> 26.71 h | Improved |
| use S3R-GS streamlined pipeline instead of 4DGF pipeline | S3R-GS · KITTI 0009 | reconstruction time (hours): 15.69 h -> 3.02 h | Improved |
| use S3R-GS streamlined pipeline instead of 4DGF pipeline | S3R-GS · KITTI 0020 | reconstruction time (hours): 18.04 h -> 3.37 h | Improved |
| use S3R-GS instead of 4DGF | S3R-GS · Argoverse 2 | PSNR: 24.97 -> 25.68 | Improved |
| use S3R-GS instead of 4DGF | S3R-GS · KITTI 0009 | PSNR: 22.64 -> 23.86 | Improved |
| use S3R-GS instead of 4DGF | S3R-GS · KITTI NVS benchmark (25%) | PSNR: 29.08 -> 29.76 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- StreetGaussians — Street Gaussians: Modeling Dynamic Urban Scenes with Gaussian Splatting (ECCV 2024) · better on 4 of 4 shared results
- EmerNeRF — EmerNeRF: Emergent Spatial-Temporal Scene Decomposition via Self-Supervision (ICLR 2024) · better on 1 of 1 shared result
- SUDS — SUDS: Scalable Urban Dynamic Scenes (CVPR 2023) · 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
- Horizon-GS: Unified 3D Gaussian Splatting for Large-Scale Aerial-to-Ground Scenes (CVPR 2025 · similarity 0.91)
- CityGaussianV2: Efficient and Geometrically Accurate Reconstruction for Large-Scale Scenes (ICLR 2025 · similarity 0.90)
- EGSRAL:An Enhanced 3D Gaussian Splatting Based Renderer with Automated Labeling for Large-Scale Driving Scene (AAAI 2025 · similarity 0.89)
- EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis (CVPR 2025 · similarity 0.89)
- Wild-GS: Real-Time Novel View Synthesis from Unconstrained Photo Collections (NeurIPS 2024 · similarity 0.89)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 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.
