Research map / Architectures and efficiency / Attention and transformers
Deep Floor Plan Recognition Using a Multi-Task Network With Room-Boundary-Guided Attention
Zhiliang Zeng, Xianzhi Li, Ying Kin Yu, Chi-Wing Fu
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 shared features and spatial contextual module (vs two separate single-task networks) (ablation) | floor plan network · R3D | overall accuracy: 0.82 -> 0.89 | Improved |
| add shared features and spatial contextual module (vs two separate single-task networks) (ablation) | floor plan network · R3D | average class accuracy: 0.72 -> 0.80 | Improved |
| add spatial contextual module (ablation) | floor plan network · R3D | overall accuracy: 0.85 -> 0.89 | Improved |
| add spatial contextual module (ablation) | floor plan network · R3D | average class accuracy: 0.72 -> 0.80 | Improved |
| add room-boundary-guided attention (ablation) | spatial contextual module · R3D | overall accuracy: 0.86 -> 0.89 | Improved |
| add room-boundary-guided attention (ablation) | spatial contextual module · R3D | average class accuracy: 0.74 -> 0.80 | Improved |
| add direction-aware kernels (ablation) | spatial contextual module · R3D | overall accuracy: 0.87 -> 0.89 | Improved |
| add direction-aware kernels (ablation) | spatial contextual module · R3D | average class accuracy: 0.77 -> 0.80 | Improved |
| add postprocessing (connected regions, majority room type) | floor plan network · R2V | overall accuracy: 0.88 -> 0.89 | Improved |
| add postprocessing to baselines DeepLabV3+ | DeepLabV3+ · R3D | overall accuracy: 0.85 -> 0.83 | Hurt |
| use floor plan network vs DeepLabV3+ | floor plan network · R3D | mean IoU: 0.50 -> 0.63 | Improved |
| use floor plan network vs RCF edge detector for walls | floor plan network · R3D | F-measure max: 0.68 -> 0.95 | Improved |
Closest papers by meaning
- FloorPlanFormer: Multi-Task Transformer Network for Floor Plan Recognition with Outer-to-Inner Feature Refinement (AAAI 2026 · similarity 0.89)
- VectorFloorSeg: Two-Stream Graph Attention Network for Vectorized Roughcast Floorplan Segmentation (CVPR 2023 · similarity 0.86)
- FloorNet: A Unified Framework for Floorplan Reconstruction from 3D Scans (ECCV 2018 · similarity 0.86)
- Tokenization Allows Multimodal Large Language Models to Understand, Generate and Edit Architectural Floor Plans (CVPR 2026 · similarity 0.83)
- Constrained Layout Generation with Factor Graphs (CVPR 2024 · similarity 0.83)
- MvFFN: Multi-view Floor-Plan Feed-Forward Network for Unposed Wide-Baseline Panorama Layout Reconstruction (NeurIPS 2026 · 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.
