Research map / Point clouds / Point cloud classification
Learning to Sample
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 task-trained S-NET samples instead of FPS | PointNet classification · ModelNet40 test, 2–256 points | classification accuracy: 18.3 -> 53.0; 35.2 -> 73.4; 61.6 -> 83.6; 76.7 -> 85.6; 82.2 -> 87.3; 86.1 -> 87.7; 87.9 -> 88.6; 88.1 -> 88.3 | Helped |
| Use task-trained S-NET samples instead of FPS | PointNet descriptor retrieval · ModelNet40 test, sampling ratios 8–64 | mAP: 58.3 -> 60.4; 49.4 -> 59.0; 37.7 -> 59.0; 27.4 -> 54.5 | Helped |
| Replace FPS with S-NET sampling | PointNet descriptor retrieval · ModelNet40 test, sampling ratios 2 and 4 | mAP: 70.1 -> 69.8; 65.7 -> 64.8 | Hurt |
| Train S-NET using a PointNet trained on FPS-sampled clouds, then retrain PointNet on S-NET samples | PointNet classification · ModelNet40 test, 32 points | classification accuracy: 82.2% -> 86.0% | Helped |
| Use S-NET rather than FPS for reconstruction sampling at high sampling ratios | Point set autoencoder · ShapeNet Core55 test, sampling ratio 32 | NRE: a little over 2 -> about 1.5 | Helped |
| Use ProgressiveNet rather than FPS for reconstruction sampling at high sampling ratios | Point set autoencoder · ShapeNet Core55 test, sampling ratio 32 | NRE: a little over 2 -> about 1.75 | Helped |
| Use ProgressiveNet instead of separately trained S-NET for sample sizes between trained sizes | PointNet classification · ModelNet40 test, intermediate sample sizes | classification accuracy | Helped |
| Remove sampling regularization (ablation) | S-NET classification sampling · ModelNet40 test, matched sampled points | classification accuracy | Hurt |
| Remove sampling regularization (ablation) | S-NET classification sampling · ModelNet40 test, generated points | classification accuracy | Helped |
| Remove the linear size-dependent factor from L_b (set δ = 0) (ablation) | S-NET matching · Input point clouds | unique matched points | Hurt |
| Remove L_b coverage-spread regularization (ablation) | ProgressiveNet classification · ModelNet40 test, sampling ratios larger than 4 and smaller than 64 | classification accuracy | Hurt |
| Remove L_m maximum-distance regularization (ablation) | S-NET classification sampling · ModelNet40 test, k=16 | classification accuracy: up to 7% decrease; 3.5-fold increase | Hurt |
Closest papers by meaning
- SAMBLE: Shape-Specific Point Cloud Sampling for an Optimal Trade-Off Between Local Detail and Global Uniformity (CVPR 2025 · similarity 0.89)
- GCD-Sampling: A General Cross-scale Decoupled Sampling for Point Cloud (AAAI 2025 · similarity 0.88)
- LTA-PCS: Learnable Task-Agnostic Point Cloud Sampling (CVPR 2024 · similarity 0.88)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.87)
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 2025 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · 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.