Research map / 3D vision / Depth and stereo
Structured Outdoor Architecture Reconstruction by Exploration and Classification
Fuyang Zhang, Xiang Xu, 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 |
|---|---|---|---|
| apply explore-and-classify on top of Conv-MPN | Conv-MPN · Aerial building reconstruction (initial Conv-MPN) | edge F1: 58.1 -> 67.1 | Improved |
| apply explore-and-classify on top of Conv-MPN | Conv-MPN · Aerial building reconstruction (initial Conv-MPN) | region F1: 54.9 -> 63.5 | Improved |
| apply explore-and-classify on top of Conv-MPN | Conv-MPN · Aerial building reconstruction (initial Conv-MPN) | corner F1: 78.8 -> 83.2 | Improved |
| apply REPL (REINFORCE-based RL) refinement to Conv-MPN | Conv-MPN · Aerial building reconstruction (initial Conv-MPN) | region F1: 54.9 -> 15.9 | Hurt |
| apply Lin et al. DQN-based refinement to Conv-MPN | Conv-MPN · Aerial building reconstruction (initial Conv-MPN) | edge F1: 58.1 -> 51.6 | Hurt |
| replace beam search with Sequential Monte Carlo (ablation) | explore-and-classify · Aerial building reconstruction (initial Conv-MPN, search comparison) | corner F1: 83.2 -> 81.2 | Mixed |
| train classifiers on Conv-MPN instead of Per-edge reconstructions, tested on Per-edge | explore-and-classify · Cross-initialization (test-time init Per-edge) | edge F1: 57.5 -> 60.3 | Improved |
| train and test on mismatched initial reconstructions (Nauata-trained, Conv-MPN test) | explore-and-classify · Cross-initialization (test-time init Conv-MPN) | edge F1: 67.1 -> 67.0 | No clear effect |
| add the geometry classification module (vs heuristic confidence-map scoring) (ablation) | explore-and-classify · Aerial building reconstruction (three initial reconstructions) | corner/edge/region F1 | Improved |
| add classifier-guided data augmentation by exploration (vs no augmentation) (ablation) | explore-and-classify · Aerial building reconstruction (Per-edge and Nauata initial) | corner/edge/region F1 | Improved |
| use classifier-guided exploration instead of random sampling for training data (ablation) | explore-and-classify · Aerial building reconstruction (Per-edge and Nauata initial) | region F1 | Improved |
| increase search depth, no-augmentation versus exploration (ablation) | explore-and-classify · Aerial building reconstruction | corner/edge/region F1 | Improved |
Closest papers by meaning
- Synthesis and Completion of Facades from Satellite Imagery (ECCV 2020 · similarity 0.86)
- Fast Building Instance Proxy Reconstruction for Large Urban Scenes (TPAMI 2024 · similarity 0.86)
- AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesis (CVPR 2025 · similarity 0.84)
- Aerial Lifting: Neural Urban Semantic and Building Instance Lifting from Aerial Imagery (CVPR 2024 · similarity 0.83)
- ArcPro: Architectural Programs for Structured 3D Abstraction of Sparse Points (CVPR 2025 · similarity 0.83)
- 3D Building Reconstruction from Monocular Remote Sensing Images with Multi-level Supervisions (CVPR 2024 · similarity 0.83)
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.
