Research map / Graph learning / Graph algorithms and clustering
Vectorizing World Buildings: Planar Graph Reconstruction by Primitive Detection and Relationship Inference
Nelson Nauata, Yasutaka Furukawa
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 corner detections to edge-only objective (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Corner f1: 53.4 -> 69.0 | Improved |
| add corner detections to edge-only objective (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Edge f1: 24.1 -> 41.2 | Improved |
| add corner-to-edge relationship constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Region f1: 12.6 -> 25.6 | Improved |
| add corner-to-edge relationship constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Edge f1: 41.2 -> 47.9 | Improved |
| add region detections and region constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Region f1: 25.6 -> 56.5 | Improved |
| add region detections and region constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Corner precision: 81.3 -> 91.7 | Improved |
| add region-to-region relationship constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Region f1: 56.5 -> 60.8 | Improved |
| add region-to-region relationship constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Edge f1: 53.6 -> 56.3 | Improved |
| add region-to-region relationship constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Region precision: 71.8 -> 70.9 | Hurt |
| add region-to-region relationship constraints (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Corner recall: 61.6 -> 64.6 | Improved |
| add corner-to-edge relationships (ablation) | IP reconstruction · Building vectorization test set (400 buildings) | Corner precision: 85.3 -> 81.3 | Hurt |
| use full IP system instead of L-CNN independent edge inference | Full system · Building vectorization test set (400 buildings) | Edge f1: 59.4 -> 56.3 | Hurt |
Closest papers by meaning
- Conv-MPN: Convolutional Message Passing Neural Network for Structured Outdoor Architecture Reconstruction (CVPR 2020 · similarity 0.88)
- Regularized Primitive Graph Learning for Unified Vector Mapping (ICCV 2023 · similarity 0.86)
- End-to-End Graph-Constrained Vectorized Floorplan Generation with Panoptic Refinement (ECCV 2022 · similarity 0.86)
- Hierarchical 3D Scene Graphs Construction Outdoors (ICCV 2025 · similarity 0.85)
- BuildingNet: Learning To Label 3D Buildings (ICCV 2021 · similarity 0.85)
- VectorFloorSeg: Two-Stream Graph Attention Network for Vectorized Roughcast Floorplan Segmentation (CVPR 2023 · 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.
