Research map / Point clouds / Segmentation and detection
Reliability-Adaptive Consistency Regularization for Weakly-Supervised Point Cloud Segmentation
Zhonghua Wu, Yicheng Wu, Guosheng Lin, Jianfei Cai
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 the full RAC-Net adaptive consistency training | KPConv baseline · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 50.1 -> 58.4 | Improved |
| use RAC-Net instead of DAT | weakly-supervised segmentation · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 56.5 -> 58.4 | Improved |
| use RAC-Net instead of DAT | weakly-supervised segmentation · S3DIS Area 5 (OTTC, 0.06% labels) | mIoU (%): 58.5 -> 60.5 | Improved |
| apply RAC-Net on a voxel-based MinkowskiNet backbone | MinkowskiNet baseline · S3DIS Area 5 (OTOC, MinkowskiNet backbone) | mIoU (%): 48.7 -> 58.6 | Improved |
| add RAC-Net built on 1T1C pseudo labels | 1T1C · ScanNet-v2 test (20 points) | mIoU (%): 59.4 -> 62.6 | Improved |
| add RAC-Net on Point Transformer backbone | Point Transformer baseline · ScanNet-v2 test (20 points) | mIoU (%): 56.4 -> 63.9 | Improved |
| add the ambiguous-set KL consistency loss L_a (ablation) | RAC-Net · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 52.5 -> 58.4 | Improved |
| add the reliable-pseudo-label mix loss L_mix (ablation) | RAC-Net · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 57.6 -> 58.4 | Improved |
| add uncertainty to confidence for dividing reliable and ambiguous sets (ablation) | RAC-Net · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 56.5 -> 57.6 | Improved |
| apply KL loss on the ambiguous set in confidence-based division (ablation) | confidence-based pseudo labeling · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 55.8 -> 56.5 | Improved |
| add adaptive division of points over uniform consistency on all points (ablation) | RAC-Net · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 56.7 -> 58.4 | Improved |
| use cross-entropy instead of Dice for the reliable set (ablation) | RAC-Net · S3DIS Area 5 (OTOC, 0.02% labels) | mIoU (%): 48.8 -> 58.4 | Improved |
Closest papers by meaning
- Dual Adaptive Transformations for Weakly Supervised Point Cloud Segmentation (ECCV 2022 · similarity 0.87)
- An MIL-Derived Transformer for Weakly Supervised Point Cloud Segmentation (CVPR 2022 · similarity 0.86)
- Weakly Supervised 3D Segmentation via Receptive-Driven Pseudo Label Consistency and Structural Consistency (AAAI 2023 · similarity 0.86)
- Weakly Supervised Semantic Point Cloud Segmentation: Towards 10x Fewer Labels (CVPR 2020 · similarity 0.86)
- Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations (AAAI 2025 · similarity 0.83)
- Distribution Guidance Network for Weakly Supervised Point Cloud Semantic Segmentation (NeurIPS 2024 · similarity 0.82)
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.
