Research map / Point clouds / Point cloud classification
Not All Neighbors Matter: Point Distribution-Aware Pruning for 3D Point Cloud
Yejin Lee, Donghyun Lee, JungUk Hong, Jae W. Lee, Hongil Yoon
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 |
|---|---|---|---|
| Prune neighborhood voxel groups with lower training-set neighbor probability, then retrain | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP32 | inference speedup: 1.60× | Improved |
| Use the lossy pruning configuration, allowing up to 1% lower accuracy metric (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP32 | inference speedup: 1.97× | Improved |
| Prune neighborhood voxel groups with lower training-set neighbor probability, then retrain | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP32 | energy efficiency improvement: 1.74× | Improved |
| Use the lossy pruning configuration, allowing up to 1% lower accuracy metric (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP32 | energy efficiency improvement: 2.22× | Improved |
| Prune neighborhood voxel groups with lower training-set neighbor probability | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP32 | FLOPs: 49.45% | Improved |
| Allow the lossy pruning configuration (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP32 | FLOPs: 62.93% | Improved |
| Replace residual blocks with Asymmetrical Residual Blocks (ablation) | MinkUNet · Four datasets, FP32 | inference speedup | Mixed |
| Replace residual blocks with Asymmetrical Residual Blocks (ablation) | MinkUNet · Four datasets | accuracy | Hurt |
| Apply neighborhood pruning to FP16 models (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP16 | inference latency improvement: 1.39× | Improved |
| Apply the lossy pruning configuration to FP16 models (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP16 | inference latency improvement: 1.61× | Improved |
| Apply neighborhood pruning to FP16 models (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP16 | energy efficiency improvement: 1.30× | Improved |
| Apply the lossy pruning configuration to FP16 models (ablation) | MinkUNet-based sparse 3D convolution models · Four datasets (geomean), FP16 | energy efficiency improvement: 1.47× | Improved |
Closest papers by meaning
- Frugal 3D Point Cloud Model Training via Progressive Near Point Filtering and Fused Aggregation (ECCV 2024 · similarity 0.91)
- FastPoint: Accelerating 3D Point Cloud Model Inference via Sample Point Distance Prediction (ICCV 2025 · similarity 0.90)
- PointCNN++: Performant Convolution on Native Points (CVPR 2026 · similarity 0.90)
- Fully-Convolutional Point Networks for Large-Scale Point Clouds (ECCV 2018 · similarity 0.88)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.88)
- LitePT: Lighter Yet Stronger Point Transformer (CVPR 2026 · similarity 0.87)
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.
