Research map / Point clouds / Shape and surface completion
Learning With Noisy Labels for Robust Point Cloud Segmentation
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 |
|---|---|---|---|
| add cluster-level voting for label correction (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8011 -> 0.8236 | Improved |
| use gamma=1 (most greedy winner label) instead of gamma=4 (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8110 vs 0.8236 | No clear effect |
| use gamma=2 instead of gamma=4 (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8209 vs 0.8236 | No clear effect |
| increase history length q to 8 (with E_warm-up=8) (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8236 -> 0.7704 | Hurt |
| use ground-truth instances instead of clustering (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8236 -> 0.8287 | Improved |
| increase E_warm-up from 5 to 8 (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8236 -> 0.7851 | Hurt |
| increase E_warm-up from 5 to 11 (ablation) | PNAL (DGCNN) · S3DIS, 20% symmetric noise | OA: 0.8569 -> 0.8343 | Hurt |
| increase E_warm-up from 5 to 11 (ablation) | PNAL (DGCNN) · S3DIS, 40% symmetric noise | OA: 0.8378 -> 0.8009 | Hurt |
| use DBSCAN eps=0.015 or 0.021 instead of 0.018 (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8206 / 0.8159 vs 0.8236 | No clear effect |
| use GMM or spectral clustering instead of DBSCAN eps=0.018 (ablation) | PNAL (DGCNN) · S3DIS, 60% symmetric noise | OA: 0.8178 / 0.8162 vs 0.8236 | No clear effect |
| add PNAL to DGCNN | DGCNN+CE · S3DIS, 60% symmetric noise | OA: 0.6390 -> 0.8236 | Improved |
| add PNAL to DGCNN under clean labels | DGCNN+CE · S3DIS, 0% noise | OA: 0.8692 -> 0.8686 | No clear effect |
Closest papers by meaning
- Robust Point Cloud Segmentation With Noisy Annotations (TPAMI 2022 · similarity 0.98)
- A Survey of Label-Efficient Deep Learning for 3D Point Clouds (TPAMI 2024 · similarity 0.89)
- 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 Upsampling Using Conditional Diffusion Module with Adaptive Noise Suppression (CVPR 2025 · similarity 0.87)
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (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.
