Research map / Point clouds / Segmentation and detection
Boundary-Aware Geometric Encoding for Semantic Segmentation of Point Clouds
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 |
|---|---|---|---|
| Use the proposed boundary-aware GEM and GCO method instead of FusionAwareConv | ScanNet v2 semantic segmentation · ScanNet v2 | mIoU: 63.0 -> 63.5 | Helped |
| Use the proposed method instead of PAT | S3DIS Area 5 (Fold #1) semantic segmentation · S3DIS Area 5 (Fold #1) | mIoU: 60.07 -> 61.43 | Helped |
| Add GCO to the PointConv baseline (ablation) | PointConv baseline · ScanNet validation | mIoU: 58.9 -> 60.4 | Helped |
| Add boundary-aware GEM without GCO (ablation) | PointConv baseline · ScanNet validation | mIoU: 58.9 -> 60.9 | Helped |
| Emphasize boundary-point features during local aggregation instead of suppressing their propagation (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 61.8 | Hurt |
| Replace GCO with single-layer Kernel Correlation using 16 kernel vectors (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 60.8 | Hurt |
| Use lightweight Kernel Correlation hierarchically with 3 vectors instead of GCO (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 61.0 | Hurt |
| Use 2 vectors in the GCO kernel instead of the proposed kernel (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 61.1 | Hurt |
| Use 6 vectors in the GCO kernel instead of the proposed kernel (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 61.7 | Hurt |
| Remove boundary information (ablation) | Proposed method · ScanNet validation | mIoU: 60.4 -> 63.4 | Hurt |
| Randomly flip 3% of predicted boundary labels (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 62.4 | Hurt |
| Exchange labels for 5% of predicted boundary points and their nearest neighbors (ablation) | Proposed method · ScanNet validation | mIoU: 63.4 -> 61.8 | Hurt |
Closest papers by meaning
- Geometry and Uncertainty-Aware 3D Point Cloud Class-Incremental Semantic Segmentation (CVPR 2023 · similarity 0.89)
- Contrastive Boundary Learning for Point Cloud Segmentation (CVPR 2022 · similarity 0.87)
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.87)
- On Geometry-Enhanced Parameter-Efficient Fine-Tuning for 3D Scene Segmentation (NeurIPS 2025 · similarity 0.87)
- S$^2$AM3D: Scale-controllable Part Segmentation of 3D Point Clouds (CVPR 2026 · similarity 0.87)
- PDF: A Probability-Driven Framework for Open World 3D Point Cloud Semantic Segmentation (CVPR 2024 · 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.