Research map / 3D vision / Gaussian splatting
Urban-GS: A Unified 3D Gaussian Splatting Framework for Compact and High-Fidelity Aerial-to-Street Reconstruction
Meng Wang, Changqun Xia, Yuze Wang, Junyi Wang, Wantong Duan, Xinxiong Xie, Yue Qi
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 Aerial-Street Joint Adaptive Densification (projected-area weighted gradients) (ablation) | Urban-GS baseline · Ablation (Elvenruin, Citysample, Road) | PSNR: 25.20 -> 25.66 | Improved |
| add AJAD (ablation) | Urban-GS baseline · Ablation (Elvenruin, Citysample, Road) | LPIPS: 0.257 -> 0.209 | Improved |
| add AJAD (ablation) | Urban-GS baseline · Ablation (Elvenruin, Citysample, Road) | anchors: 2774k -> 9713k | Hurt |
| add Contribution-based Anchor Pruning (ablation) | Urban-GS + AJAD · Ablation (Elvenruin, Citysample, Road) | anchors: 9713k -> 2785k | Improved |
| add Contribution-based Anchor Pruning (ablation) | Urban-GS + AJAD · Ablation (Elvenruin, Citysample, Road) | PSNR: 25.66 -> 25.50 | Hurt |
| add Global-to-Local Optimization (ablation) | Urban-GS + AJAD + CAP · Ablation (Elvenruin, Citysample, Road) | PSNR: 25.50 -> 26.05 | Improved |
| replace GLO with 20k extra uniform-sampling iterations (ablation) | Urban-GS + AJAD + CAP · Ablation (Elvenruin, Citysample, Road) | PSNR: 25.50 -> 25.59 | No clear effect |
| use projected-area weighted densification vs Abs-GS densification (ablation) | Base 3DGS-style densification · Ablation (Elvenruin, Citysample, Road) | PSNR: 25.25 (Abs-GS) vs 25.66 (ours) | Improved |
| use projected-area weighted densification vs Hier-GS max-gradient (ablation) | Base 3DGS-style densification · Ablation (Elvenruin, Citysample, Road) | PSNR: 25.47 (Hier-GS) vs 25.66 (ours) | Improved |
| use contribution-weighted mask regularization vs global MaskGaussian loss (lambda_m 0.003) (ablation) | Anchor pruning · Ablation (Elvenruin, Citysample, Road) | PSNR: 24.99 -> 25.50 | Improved |
| accumulate densification gradients from merged aerial+street views instead of single view type (ablation) | 3DGS densification · Colosseum | PSNR (aerial test): 25.64 (aerial only) -> 25.35 (merge) | Hurt |
| use Urban-GS instead of Horizon-GS on UC-GS dataset | Urban-GS · UC-GS View(+1m5°down) | PSNR: 25.37 -> 25.23 PSNR; SSIM 0.761 -> 0.765 | Mixed |
Closest papers by meaning
- Horizon-GS: Unified 3D Gaussian Splatting for Large-Scale Aerial-to-Ground Scenes (CVPR 2025 · similarity 0.94)
- CityGaussianV2: Efficient and Geometrically Accurate Reconstruction for Large-Scale Scenes (ICLR 2025 · similarity 0.91)
- EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis (CVPR 2025 · similarity 0.91)
- D$^2$GS: Dense Depth Regularization for LiDAR-free Urban Scene Reconstruction (NeurIPS 2025 · similarity 0.90)
- PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- AtlasGS: Atlanta-world Guided Surface Reconstruction with Implicit Structured Gaussians (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.
