Research map / Point clouds / Segmentation and detection
Attention Discriminant Sampling for Point Clouds
Cheng-Yao Hong, Yu-Ying Chou, Tyng-Luh Liu
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 |
|---|---|---|---|
| use ADS instead of prior best on ModelNet40 with normals | ADS · ModelNet40 | overall accuracy (%): 94.4 -> 95.1 | Improved |
| use ADS instead of PointNeXt-S on ScanObjectNN | ADS · ScanObjectNN | overall accuracy (%): 87.7 -> 87.5 | Hurt |
| use ADS instead of PointTransformer-V2 on S3DIS Area-5 | ADS · S3DIS Area-5 | mIoU (%): 71.6 -> 71.8 | No clear effect |
| use ADS instead of prior SOTA on S3DIS 6-fold | ADS · S3DIS 6-fold | mIoU (%): 74.3 -> 75.1 | Improved |
| replace backbone with ADS in GroupFree | GroupFree · ScanNetV2 detection | mAP@0.5 (%): 54.8 -> 55.0 | Improved |
| add the cluster-based formulation (ablation) | ADS · ModelNet40 | overall accuracy (%): +1.1% OA | Improved |
| add EMD reconstruction loss (ablation) | ADS · ModelNet40 | overall accuracy (%) and robustness to sampling ratio | Improved |
| add relative normal angle to attention bias (ablation) | ADS · ModelNet40 | overall accuracy (%): 94.1 -> 94.4 | Improved |
| add MLP-predicted positional bias instead of none (ablation) | ADS · ModelNet40 | overall accuracy (%): 93.9 -> 94.4 | Improved |
| use larger scaling ratio for k in ADS modules (k=8, [1,2,4] vs [1,2,2]) (ablation) | ADS · ModelNet40 | overall accuracy (%): 94.5 -> 93.6 | Hurt |
| reduce sampling ratio (increase 1/r) (ablation) | ADS vs other samplers · ModelNet40 classification and reconstruction | performance degradation | Improved |
Closest papers by meaning
- Unsupervised Point Cloud Object Co-Segmentation by Co-Contrastive Learning and Mutual Attention Sampling (ICCV 2021 · similarity 0.89)
- Attentional ShapeContextNet for Point Cloud Recognition (CVPR 2018 · similarity 0.89)
- Point Transformer (ICCV 2021 · similarity 0.89)
- Clustering-based Supervised Training for 3D Perception (TPAMI 2026 · similarity 0.89)
- Efficient Outdoor 3D Point Cloud Semantic Segmentation for Critical Road Objects and Distributed Contexts (ECCV 2020 · similarity 0.88)
- SAMBLE: Shape-Specific Point Cloud Sampling for an Optimal Trade-Off Between Local Detail and Global Uniformity (CVPR 2025 · 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.
