Research map / Point clouds / Point cloud classification
Modify Self-Attention via Skeleton Decomposition for Effective Point Cloud Transformer
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 |
|---|---|---|---|
| Replace standard self-attention with SD-SA (|Γ|=64) | SPCT · ShapeNet | mIoU (%): 85.6 -> 85.5 | No clear effect |
| Replace standard self-attention with SD-SA | SPCT · ModelNet40 | mAcc (%): 92.0 -> 91.8 | No clear effect |
| Add a grouping layer (ablation) | SD-SA on ModelNet40 · ModelNet40 | mAcc (%): 91.8 -> 92.2 | Helped |
| Replace the PointNet detector with an SD-based transformer | LiDAR-RCNN · KITTI validation (Car) | Easy: 86.8 -> 86.7 | No clear effect |
| Replace the PointNet detector with an SD-based transformer | LiDAR-RCNN · KITTI validation (Car) | Moderate: 79.0 -> 78.7 | No clear effect |
| Replace the PointNet detector with an SD-based transformer | LiDAR-RCNN · KITTI validation (Car) | Hard: 70.4 -> 70.2 | No clear effect |
| Use l1-norm row selection rather than FPS (ablation) | SD-SA · ShapeNet | mIoU (%): 85.2 -> 85.6 | Helped |
| Use l1-norm row selection rather than mean attention from three random samples (ablation) | SD-SA · ShapeNet | mIoU (%): 85.3 -> 85.6 | Helped |
| Use l1-norm row selection rather than random selection (ablation) | SD-SA · ShapeNet | mIoU (%): 85.4 -> 85.6 | No clear effect |
| Replace one standard self-attention layer with SD-SA (ablation) | SPCT · ShapeNet | mIoU (%): 85.6 -> 85.6 | No clear effect |
| Replace two standard self-attention layers with SD-SA (ablation) | SPCT · ShapeNet | mIoU (%): 85.6 -> 85.5 | No clear effect |
| Replace three standard self-attention layers with SD-SA (ablation) | SPCT · ShapeNet | mIoU (%): 85.6 -> 85.5 | No clear effect |
Closest papers by meaning
- Self-Positioning Point-Based Transformer for Point Cloud Understanding (CVPR 2023 · similarity 0.89)
- FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformer (CVPR 2023 · similarity 0.89)
- PatchFormer: An Efficient Point Transformer With Patch Attention (CVPR 2022 · similarity 0.87)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.86)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.86)
- Scene Flow Prior Based Point Cloud Completion with Masked Transformer (Student Abstract) (AAAI 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.