Research map / Point clouds / Point cloud classification
VBA: Vector Bundle Attention for Intrinsically Geometric Representation Learning
Shenglei Fang, Xianfang Sun, You Zhou
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 |
|---|---|---|---|
| replace standard attention with vector bundle attention (ablation) | VBA-T · ModelNet40 | overall accuracy: 82.6% -> 92.9% | Improved |
| replace standard attention with vector bundle attention (ablation) | VBA-T · PBMC | overall accuracy: 66.76% -> 78.71% | Improved |
| transport-then-attend geometry instead of additive geometric bias (GBT) (ablation) | VBA-T · PBMC | overall accuracy: 70.32% -> 78.71% | Improved |
| transport-then-attend geometry instead of additive geometric bias (GBT) (ablation) | VBA-T · ModelNet40 | overall accuracy: 85.8% -> 92.9% | Improved |
| use VBA instead of efficient MQA or GQA attention variants (ablation) | VBA-T · Pancreas | F1: 0.67 (MQA-T), 0.66 (GQA-T) -> 0.86 | Improved |
| add self-supervised pretraining on HCL (ablation) | VBA-SC · PBMC | overall accuracy: 78.71% -> 80.68% | Improved |
| add self-supervised pretraining on HCL (ablation) | VBA-SC · Pancreas | overall accuracy: 93.89% -> 97.64% | Improved |
| pretrain scGPT and scBERT on same HCL subset as VBA-SC (ablation) | scGPT, scBERT · PBMC | overall accuracy: scGPT 75.47% -> 74.42%; scBERT 75.52% -> 74.16% | Hurt |
| restrict attention to top-k neighbors on learned manifold (ablation) | VBA-SC · PBMC | accuracy: 78.71 -> 77.65 accuracy; 876 -> 560 s/epoch | Mixed |
| use low-rank transport generators (ablation) | VBA-SC · PBMC | accuracy: 78.71 -> 76.90 accuracy; 876 -> 529 s/epoch | Mixed |
| use VBA-P on point clouds vs specialized models | VBA-P · ModelNet40 | overall accuracy: 92.9% vs 93.8% (PointGA) | No clear effect |
| VBA-ST vs graph-based spatial clustering methods | VBA-ST · DLPFC 12 slices | average ARI: 0.45 (SpaGCN), 0.43 (GraphST) -> 0.50 | Improved |
Closest papers by meaning
- GTA: A Geometry-Aware Attention Mechanism for Multi-View Transformers (ICLR 2024 · similarity 0.84)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.84)
- Partition First, Embed Later: Laplacian-Based Feature Partitioning for Refined Embedding and Visualization of High-Dimensional Data (ICML 2025 · similarity 0.84)
- Graph Geometry-Preserving Autoencoders (ICML 2024 · similarity 0.84)
- A Linear Time and Space Local Point Cloud Geometry Encoder via Vectorized Kernel Mixture (VecKM) (ICML 2024 · similarity 0.83)
- Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves (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.
