Research map / Point clouds / Segmentation and detection
Learning Regional Purity for Instance Segmentation on 3D Point Clouds
Shichao Dong, Guosheng Lin, Tzu-Yi Hung
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 |
|---|---|---|---|
| add extra grouping radius for high purity points (ablation) | baseline clustering · ScanNet v2 val | mAP: 0.284 -> 0.294 | Improved |
| add extra grouping radius for high purity points (ablation) | baseline clustering · ScanNet v2 val | mAP@0.5: 0.508 -> 0.518 | Improved |
| add low purity points as soft barrier grouped only by medium purity points (ablation) | clustering with high purity tolerance · ScanNet v2 val | mAP: 0.294 -> 0.322 | Improved |
| add low purity points as soft barrier grouped only by medium purity points (ablation) | clustering with high purity tolerance · ScanNet v2 val | mAP@0.5: 0.518 -> 0.545 | Improved |
| shift high purity points by predicted offset (ablation) | clustering with purity guidance · ScanNet v2 val | mAP: 0.322 -> 0.352 | Improved |
| shift high purity points by predicted offset (ablation) | clustering with purity guidance · ScanNet v2 val | mAP@0.5: 0.545 -> 0.572 | Improved |
| add offset direction cosine similarity criterion (ablation) | clustering with offset shift · ScanNet v2 val | mAP: 0.352 -> 0.359 | Improved |
| add offset direction cosine similarity criterion (ablation) | clustering with offset shift · ScanNet v2 val | mAP@0.5: 0.572 -> 0.582 | Improved |
| pretrain on toy scenes instead of training from scratch (ablation) | RPGN · ScanNet v2 val | mAP: 3.8% mAP | Improved |
| use refined semantic prediction via label smoothing | RPGN · ScanNet v2 val | mAP@0.5 (avg): 61.9 -> 64.2 | Improved |
| use pretrained purity model, evaluate on S3DIS 6-fold instead of Area 5 | RPGN · S3DIS | mPrec: 64.0 -> 84.5 | Improved |
| replace PointGroup dual-set proposals plus NMS with single-set purity-guided clustering | RPGN · ScanNet v2 val | total processing time (s): 141 -> 89 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- PointGroup — PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation (CVPR 2020) · better on 3 of 3 shared results
- 3D-MPA — 3D-MPA: Multi-Proposal Aggregation for 3D Semantic Instance Segmentation (CVPR 2020) · better on 1 of 1 shared result
- SGPN — SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation (CVPR 2018) · 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
- Divide and Conquer: 3D Point Cloud Instance Segmentation With Point-Wise Binarization (ICCV 2023 · similarity 0.88)
- PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation (CVPR 2020 · similarity 0.88)
- PointInst3D: Segmenting 3D Instances by Points (ECCV 2022 · similarity 0.87)
- ISBNet: A 3D Point Cloud Instance Segmentation Network With Instance-Aware Sampling and Box-Aware Dynamic Convolution (CVPR 2023 · similarity 0.87)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.86)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.86)
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.
