Research map / Point clouds / Segmentation and detection
SensatUrban: Learning Semantics from Urban-Scale Photogrammetric Point Clouds
Qingyong Hu, Bo Yang, Sheikh Khalid, Wen Xiao, Niki Trigoni, Andrew Markham
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 RGB color input (ablation) | PointNet · SensatUrban | mIoU (%): 28.85 -> 49.69 | Improved |
| add RGB color input (ablation) | PointNet++ · SensatUrban | mIoU (%): 50.71 -> 58.13 | Improved |
| add RGB color input (ablation) | SPGraph · SensatUrban | mIoU (%): 35.29 -> 37.29 | Improved |
| add RGB color input (ablation) | KPConv · SensatUrban | mIoU (%): 51.79 -> 64.50 | Improved |
| add RGB color input (ablation) | RandLA-Net · SensatUrban | mIoU (%): 51.53 -> 58.14 | Improved |
| use constant-number instead of constant-volume input sets (grid sampling) (ablation) | RandLA-Net · SensatUrban | mIoU (%): 49.18 -> 58.64 | Improved |
| use grid instead of random downsampling (constant number) (ablation) | RandLA-Net · SensatUrban | mIoU (%): 57.55 -> 58.64 | No clear effect |
| add weighted cross-entropy with sqrt weights (ablation) | RandLA-Net · SensatUrban | mIoU (%): 57.77 -> 62.80 | Improved |
| replace cross-entropy with Lovasz loss (ablation) | PointNet · SensatUrban | mIoU (%): 49.69 -> 52.53 | Improved |
| replace cross-entropy with focal loss (ablation) | RandLA-Net · SensatUrban | mIoU (%): 57.77 -> 60.41 | Improved |
| train on 1% labels per category instead of full labels | RandLA-Net · SensatUrban | mIoU (%): 58.14 -> 55.80 | No clear effect |
| test on another city (cross-city) instead of the training city | KPConv trained on Birmingham · SensatUrban | mIoU (%): 61.65 -> 40.51 | Hurt |
Closest papers by meaning
- Towards Semantic Segmentation of Urban-Scale 3D Point Clouds: A Dataset, Benchmarks and Challenges (CVPR 2021 · similarity 0.93)
- CityRefer: Geography-aware 3D Visual Grounding Dataset on City-scale Point Cloud Data (NeurIPS 2023 · similarity 0.90)
- USCILab3D: A Large-scale, Long-term, Semantically Annotated Outdoor Dataset (NeurIPS 2024 · similarity 0.89)
- SUM Parts: Benchmarking Part-Level Semantic Segmentation of Urban Meshes (CVPR 2025 · similarity 0.86)
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.86)
- Learnable Earth Parser: Discovering 3D Prototypes in Aerial Scans (CVPR 2024 · similarity 0.86)
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.
