Research map / Point clouds / Segmentation and detection
Graph Attention Convolution for Point Cloud Semantic Segmentation
Check 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 GAC attention instead of max aggregation (ablation) | GACNet · S3DIS Area 5 | mIoU: 58.42 -> 62.85; 4.43% higher | Helped |
| Use both spatial positions and feature attributes to generate attention (ablation) | GACNet · S3DIS Area 5 | mIoU: 60.41 -> 62.85; 60.25 -> 62.85 | Helped |
| Replace the final GAC layer with CRF-RNN using 1 iteration (ablation) | GACNet · S3DIS Area 5 | mIoU: 61.70 -> 62.85 | Hurt |
| Replace the final GAC layer with CRF-RNN using 3 iterations (ablation) | GACNet · S3DIS Area 5 | OA: 87.86 -> 87.79 | Mixed |
| Replace the final GAC layer with CRF-RNN using 5 iterations (ablation) | GACNet · S3DIS Area 5 | mIoU: 61.83 -> 62.85 | Hurt |
| Keep RGB in the initial point features (ablation) | GACNet · S3DIS Area 5 | mIoU: 60.16 -> 62.85; drops by 2.69% without RGB | Helped |
| Keep geo-feature in the initial point features (ablation) | GACNet · S3DIS Area 5 | mIoU: 60.37 -> 62.85; drops by 2.48% without geo-feature | Helped |
| Keep both RGB and geo-feature in the initial point features (ablation) | GACNet · S3DIS Area 5 | mIoU: drops by 3.89% without both | Helped |
| Use GACNet rather than SPG | point cloud segmentation · S3DIS Area 5 | mIoU: 58.04 -> 62.85 | Helped |
| Use GACNet rather than SPG | point cloud segmentation · Semantic3D reduced-8 | mIoU: 73.2 -> 70.8 | Hurt |
| Use GACNet rather than SPG | point cloud segmentation · Semantic3D reduced-8 | OA: 94.0 -> 91.9 | Hurt |
Closest papers by meaning
- Improving Graph Representation for Point Cloud Segmentation via Attentive Filtering (CVPR 2023 · similarity 0.89)
- Exploiting Local and Global Structure for Point Cloud Semantic Segmentation with Contextual Point Representations (NeurIPS 2019 · similarity 0.89)
- Large-Scale Point Cloud Semantic Segmentation With Superpoint Graphs (CVPR 2018 · similarity 0.87)
- Attentional ShapeContextNet for Point Cloud Recognition (CVPR 2018 · similarity 0.87)
- DHGCN: Dynamic Hop Graph Convolution Network for Self-Supervised Point Cloud Learning (AAAI 2024 · similarity 0.85)
- Taylor Series-Inspired Local Structure Fitting Network for Few-shot Point Cloud Semantic Segmentation (AAAI 2025 · similarity 0.84)
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.