Research map / Point clouds / Segmentation and detection
Robust Point Cloud Segmentation With Noisy Annotations
Shuquan Ye, Dongdong Chen, Songfang Han, Jing Liao
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 PNAL instead of plain CE | DGCNN · S3DIS 60% symmetric noise | OA: 0.6390 -> 0.8236 | Improved |
| use PNAL instead of SELFIE | DGCNN · S3DIS 60% symmetric noise | OA: 0.7725 -> 0.8236 | Improved |
| replace CE with SCE | DGCNN · S3DIS 60% symmetric noise | OA: 0.6390 -> 0.6509 | No clear effect |
| replace CE with GCE | DGCNN · S3DIS 0% noise | OA: 0.8692 -> 0.7067 | Hurt |
| add PNAL on clean data | DGCNN · S3DIS 0% noise | OA: 0.8692 -> 0.8686 | No clear effect |
| add PNAL-boundary over PNAL | DGCNN · S3DIS boundary noise (alpha=1.0, beta=0.7) | OA: 0.6992 -> 0.8536 | Improved |
| add PNAL on re-labeled ScanNetV2 | SparseConvNet · ScanNetV2 re-labeled val | mIoU: 0.7103 -> 0.7416 | Improved |
| add PNAL on real noisy ScanNetV2 val | SparseConvNet · ScanNetV2 noisy val | mIoU: 0.7250 -> 0.7298 | No clear effect |
| add cluster-level voting (ablation) | PNAL · S3DIS 60% symmetric noise | OA: 0.8011 -> 0.8236 | Improved |
| use gamma=4 instead of gamma=1 (ablation) | PNAL · S3DIS 60% symmetric noise | OA: 0.8110 -> 0.8236 | No clear effect |
| use history length 5 instead of 8 (warm-up 5 instead of 8) (ablation) | PNAL · S3DIS 60% symmetric noise | OA: 0.7704 -> 0.8236 | Improved |
| shorten warm-up from 11 to 5 epochs (ablation) | PNAL · S3DIS 60% symmetric noise | OA: 0.7812 -> 0.8236 | Improved |
Closest papers by meaning
- Learning With Noisy Labels for Robust Point Cloud Segmentation (ICCV 2021 · similarity 0.98)
- Weakly Supervised Semantic Point Cloud Segmentation: Towards 10x Fewer Labels (CVPR 2020 · similarity 0.88)
- A Survey of Label-Efficient Deep Learning for 3D Point Clouds (TPAMI 2024 · similarity 0.88)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.87)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.87)
- Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations (AAAI 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.
