Research map / Point clouds / Segmentation and detection
Hierarchical Point-based Active Learning for Semi-supervised Point Cloud Semantic Segmentation
Zongyi Xu, Bo Yuan, Shanshan Zhao, Qianni Zhang, Xinbo Gao
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 random point selection with hierarchical point-based active selection (HMMU + FDS) (ablation) | Point-based active learning, MinkowskiNet · S3DIS Area 5 (0.1% labels) | mIoU: 54.7 -> 59.3 | Improved |
| add hierarchical voxel-level contextual uncertainty (HMMU) to point-level MMU (ablation) | Active learning module · S3DIS Area 5 (0.1% labels) | mIoU: 55.51 -> 57.65 | Improved |
| add feature-distance suppression (FDS) to MMU selection (ablation) | Active learning module · S3DIS Area 5 (0.1% labels) | mIoU: 55.51 -> 56.68 | Improved |
| add teacher-student semi-supervised module on top of HMMU + FDS (ablation) | Hierarchical point-based active learning · S3DIS Area 5 (0.1% labels) | mIoU: 59.29 -> 62.26 | Improved |
| use entropy or least-confidence point selection instead of random | Point-based active learning · S3DIS Area 5 (0.1% labels) | mIoU: 54.7 -> 48.2 (Entropy), 53.3 (LC) | Hurt |
| use point-based selection instead of region-based selection to reduce labelling | Active learning · S3DIS Area 5 | labelled points for 90% of fully-supervised: 15% -> 0.1% | Improved |
| use one point-level and three voxel-level layers in HMMU (ablation) | HMMU · S3DIS Area 5 (0.1% labels) | mIoU iter 5: 56.68 -> 59.29 | Improved |
| add a fourth voxel-level layer in HMMU (v_r=200cm) (ablation) | HMMU · S3DIS Area 5 (0.1% labels) | mIoU iter 5: 59.29 -> 59.19 | No clear effect |
| set FDS radius 20cm and threshold 0.8 rather than other combinations (ablation) | FDS · S3DIS Area 5 (0.1% labels) | mIoU iter 5: 58.79 -> 59.29 | Improved |
| use proposed method with 20 labelled points per scene versus GaIA on ScanNetV2 | Hierarchical point-based active learning · ScanNetV2 test (20 pts per scene) | mIoU: 63.8 vs 62.5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ReDAL — ReDAL: Region-Based and Diversity-Aware Active Learning for Point Cloud Semantic Segmentation (ICCV 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · similarity 0.91)
- PARI: Policy-Driven Active Residual Intervention for Weakly Supervised Point Cloud Segmentation (NeurIPS 2026 · similarity 0.90)
- Multi-Path Region Mining for Weakly Supervised 3D Semantic Segmentation on Point Clouds (CVPR 2020 · similarity 0.90)
- Less Is More: Label Recommendation for Weakly Supervised Point Cloud Semantic Segmentation (AAAI 2024 · similarity 0.89)
- Multi-Modality Affinity Inference for Weakly Supervised 3D Semantic Segmentation (AAAI 2024 · similarity 0.88)
- Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.88)
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.
