Research map / Point clouds / Segmentation and detection
High-Quality Pseudo-Labeling for Point Cloud Segmentation With Scene-Level Annotation
Lunhao Duan, Shanshan Zhao, Xingxing Weng, Jing Zhang, Gui-Song Xia
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 cross-modal feature guidance (3D contrastive loss) (ablation) | baseline · ScanNet train pseudo-labels 3D | mIoU: 25.1 -> 36.8 | Improved |
| add region-point consistency without CMG (ablation) | baseline · ScanNet train pseudo-labels 3D | mIoU: 25.1 -> 29.6 | Improved |
| add region-point consistency (3D) on top of CMG (ablation) | CMG · ScanNet train pseudo-labels 3D | mIoU: 36.8 -> 47.7 | Improved |
| add 2D branch region-point consistency loss (ablation) | CMG + RPC 3D · ScanNet train pseudo-labels 3D | mIoU: 47.7 -> 48.2 | Improved |
| add Mix3D augmentation (ablation) | full method · ScanNet train pseudo-labels 3D | mIoU: 48.2 -> 50.1 | Improved |
| add Mix3D augmentation (ablation) | full method · S3DIS train pseudo-labels 3D | mIoU: 50.5 -> 51.6 | Improved |
| use contrastive distillation instead of MSE for 2D-to-3D (ablation) | CMG · ScanNet train pseudo-labels 3D | mIoU: 32.6 -> 36.8 | Improved |
| use contrastive distillation instead of CAM guidance (ablation) | CMG · ScanNet train pseudo-labels 3D | mIoU: 32.7 -> 36.8 | Improved |
| use unsupervised regions instead of point-wise consistency (ablation) | RPC · ScanNet train pseudo-labels | mIoU: 44.5 -> 48.2 | Improved |
| use adaptive per-category threshold instead of fixed (eta 0.8) (ablation) | RPC · ScanNet train pseudo-labels | mIoU: 46.6 -> 48.2 | Improved |
| use 6 image views instead of none, versus 12 views (ablation) | CMG · ScanNet train pseudo-labels 3D | mIoU: 25.1 -> 36.8; 36.8 -> 33.1 | Mixed |
| initialize 2D backbone with ImageNet weights (ablation) | ResNet50 · ScanNet train pseudo-labels 3D | mIoU: 40.0 -> 48.2 | Improved |
Closest papers by meaning
- Quantity-Quality Enhanced Self-Training Network for Weakly Supervised Point Cloud Semantic Segmentation (TPAMI 2025 · similarity 0.91)
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · similarity 0.91)
- Multi-Path Region Mining for Weakly Supervised 3D Semantic Segmentation on Point Clouds (CVPR 2020 · similarity 0.90)
- Weakly Supervised 3D Segmentation via Receptive-Driven Pseudo Label Consistency and Structural Consistency (AAAI 2023 · 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 0.89)
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.
