Research map / Point clouds / Segmentation and detection
An MIL-Derived Transformer for Weakly Supervised Point Cloud Segmentation
Cheng-Kun Yang, Ji-Jia Wu, Kai-Syun Chen, Yung-Yu Chuang, Yen-Yu Lin
Check 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 the proposed method instead of WYPR under scene-level supervision | Weakly supervised point-cloud segmentation · S3DIS scene-level | mIoU: 22.3 -> 12.9 | Hurt |
| Use the proposed method instead of MPRM under subcloud-level supervision | Weakly supervised point-cloud segmentation · ScanNet | mIoU: 43.2 -> 47.4; 41.1 -> 45.8 | Improved |
| Use the proposed method instead of OTOC† with 20 labeled points | Weakly supervised point-cloud segmentation · ScanNet test | mIoU: 55.1 -> 54.4 | Hurt |
| Use the proposed method instead of PSD with 0.02% labeled points | Weakly supervised point-cloud segmentation · S3DIS test | mIoU: 48.2 -> 51.4 | Improved |
| Add the MIL-derived transformer loss (ablation) | CAM classification baseline · ScanNet and S3DIS, sparse labels | mIoU: 52.3 -> 55.4; 46.3 -> 49.1 | Improved |
| Add cross-scale consistency to the MIL model (ablation) | MIL-derived transformer · ScanNet and S3DIS, sparse labels | mIoU: 55.4 -> 55.9; 49.1 -> 49.6 | Improved |
| Add AdaGWP to MIL and cross-scale consistency (ablation) | MIL-derived transformer · ScanNet and S3DIS, sparse labels | mIoU: 55.9 -> 57.8; 49.6 -> 51.4 | Improved |
| Add an ordinary transformer encoder without the MIL formulation (ablation) | CAM classification baseline · ScanNet and S3DIS, sparse labels | mIoU: 52.3 -> 55.0; 46.3 -> 47.9 | Improved |
| Use only positive MIL bags rather than both bag types (ablation) | MIL-derived transformer · ScanNet and S3DIS, sparse labels | mIoU: 57.8 -> 57.4; 51.4 -> 49.1 | Hurt |
| Increase the relative weight of negative bags (α=0.3, β=0.7) (ablation) | MIL-derived transformer · ScanNet and S3DIS, sparse labels | mIoU: 57.8 -> 57.1; 51.4 -> 51.2 | Hurt |
| Use only negative MIL bags rather than both bag types (ablation) | MIL-derived transformer · ScanNet and S3DIS, sparse labels | mIoU: 57.8 -> 56.9; 51.4 -> 48.9 | Hurt |
| Increase the cross-scale sampling ratio from γ=0.25 to γ=0.5 (ablation) | MIL-derived transformer with cross-scale consistency · ScanNet, sparse labels | mIoU: 55.5 -> 56.4 | Improved |
Closest papers by meaning
- Dual Adaptive Transformations for Weakly Supervised Point Cloud Segmentation (ECCV 2022 · similarity 0.91)
- Distribution Guidance Network for Weakly Supervised Point Cloud Semantic Segmentation (NeurIPS 2024 · similarity 0.88)
- WeakPCSOD: Overcoming the Bias of Box Annotations for Weakly Supervised Point Cloud Salient Object Detection (AAAI 2024 · similarity 0.87)
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · similarity 0.87)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (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.
