Research map / Point clouds / Segmentation and detection
BuildingNet: Learning To Label 3D Buildings
Pratheba Selvaraju, Mohamed Nabail, Marios Loizou, Maria Maslioukova, Melinos Averkiou, Andreas Andreou, Siddhartha Chaudhuri, Evangelos Kalogerakis
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 the building GNN over subgroup graph on MinkowskiNet features | MinkNet2Sub · BuildingNet-Mesh (no color) | Part IoU (%): 33.1 -> 40.0 | Improved |
| add the building GNN over subgroup graph on MinkowskiNet features | MinkNet2Sub · BuildingNet-Mesh (with color) | Part IoU (%): 37.0 -> 42.6 | Improved |
| add the building GNN over subgroup graph on PointNet++ features | PointNet2Sub · BuildingNet-Mesh (with color) | Part IoU (%): 16.1 -> 31.5 | Improved |
| replace graph cuts with the GNN | MinkNet-GC · BuildingNet-Mesh (no color) | Part IoU (%): 29.9 -> 40.0 | Improved |
| aggregate point predictions per mesh subgroup instead of per triangle | MinkNet · BuildingNet-Mesh (no color) | Part IoU (%): 28.8 -> 33.1 | Improved |
| add RGB color input | MinkNet · BuildingNet-Points | Part IoU (%): 26.9 -> 29.9 | Improved |
| use average pooling instead of max pooling over subgroup points (ablation) | MinkNet2Sub · BuildingNet-Mesh (no color) | Part IoU (%): 30.4 -> 33.1 | Improved |
| use average pooling instead of max pooling over triangle points (ablation) | MinkNet2Triangle · BuildingNet-Mesh (with color) | Part IoU (%): 31.5 -> 32.8 | Improved |
| use weighted cross-entropy instead of unweighted cross-entropy (ablation) | MinkNet · BuildingNet-Points (no color) | Part IoU (%): 24.5 -> 26.9 | Improved |
| use weighted cross-entropy instead of alpha-balanced focal loss (ablation) | MinkNet · BuildingNet-Points (no color) | Part IoU (%): 22.3 -> 26.9 | Improved |
| use class-balanced loss instead of weighted cross-entropy (ablation) | MinkNet · BuildingNet-Points (with color) | Part IoU (%): 29.9 -> 30.4 | No clear effect |
| add each edge type (proximity, support, similarity, containment) to node features (ablation) | BuildingGNN · BuildingNet-Mesh | Part IoU (%) | Improved |
Closest papers by meaning
- Intelligent Home 3D: Automatic 3D-House Design From Linguistic Descriptions Only (CVPR 2020 · similarity 0.87)
- 3D Semantic Parsing of Large-Scale Indoor Spaces (CVPR 2016 · similarity 0.87)
- Building3D: A Urban-Scale Dataset and Benchmarks for Learning Roof Structures from Point Clouds (ICCV 2023 · similarity 0.87)
- ArchiSet: Benchmarking Editable and Consistent Single-View 3D Reconstruction of Buildings with Specific Window-to-Wall Ratios (ICCV 2025 · similarity 0.87)
- BuildingWorld: A Structured 3D Building Dataset for Urban Foundation Models (AAAI 2026 · similarity 0.87)
- HouseLayout3D: A Benchmark and Training-free Baseline for 3D Layout Estimation in the Wild (NeurIPS 2025 · similarity 0.85)
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.
