Research map / Detection and segmentation / CNN features
VectorFloorSeg: Two-Stream Graph Attention Network for Vectorized Roughcast Floorplan Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 image (CNN) features from rasterized floorplan to vertex embedding (ablation) | two-stream GNN vertex embedding · R2V | mIoU: +about 13.7 (45.92 to 59.01 with pos enc) | Helped |
| add interior-sampled dual vertex embedding (Eq. 3) instead of centroid only (ablation) | dual vertex embedding · R2V | mIoU: mIoU 77.08 vs 75.60 with sampling added to the full stack | Helped |
| add primal stream (boundary line classification) plus modulated GAT layer (ablation) | two-stream GNN · R2V | mIoU: +4.17 mIoU, +2.97 mAcc, +3.64 RI | Helped |
| use modulated GAT (dual-edge feature modulation) instead of vanilla similarity attention (ablation) | GAT layers in two-stream GNN · R2V | mIoU: 79.77 vs 78.22 (with dual map and edge init) | Helped |
| add edge embeddings (Eqs. 4 and 5) for primal and dual edges (ablation) | two-stream GNN edge features · R2V | mIoU: 79.77 vs 77.38 | Helped |
| use modulation by dual edge features from the other stream in GAT layers (ablation) | modulated GAT · R2V | mIoU: 79.77 vs 78.22 | Helped |
| replace image-based segmentation (DeepLabV3+, OCRNet, DFPR) with vector two-stream GAT | roughcast floorplan segmentation · R2V and CubiCasa-5k | mIoU: R2V 79.77 (ResNet-50 variant) vs 74.59 DeepLabV3+; CubiCasa test 57.48 vs 58.18 DeepLabV3+ (ResNet-50 variant 62.49) | Helped |
| vector two-stream GAT vs image-based methods on RI metric | roughcast floorplan segmentation · R2V and CubiCasa-5k | RI: R2V 84.67 vs 70.82 (OCRNet); CubiCasa test 64.09 vs 44.71 (UperNet) | Helped |
Closest papers by meaning
- GSDiff: Synthesizing Vector Floorplans via Geometry-enhanced Structural Graph Generation (AAAI 2025 · similarity 0.88)
- CAGE: Continuity-Aware edGE Network Unlocks Robust Floorplan Reconstruction (NeurIPS 2025 · similarity 0.86)
- MvFFN: Multi-view Floor-Plan Feed-Forward Network for Unposed Wide-Baseline Panorama Layout Reconstruction (NeurIPS 2026 · similarity 0.85)
- SegGraph: Leveraging Graphs of SAM Segments for Few-Shot 3D Part Segmentation (NeurIPS 2025 · similarity 0.84)
- Constrained Layout Generation with Factor Graphs (CVPR 2024 · similarity 0.84)
- Seg2Reg: Differentiable 2D Segmentation to 1D Regression Rendering for 360 Room Layout Reconstruction (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.