Research map / Point clouds / Segmentation and detection
Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes
Julian Chibane, Francis Engelmann, Tuan Anh Tran, Gerard Pons-Moll
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 |
|---|---|---|---|
| supervise undecided points with the smallest/closest box (ablation) | Box2Mask · ScanNet val | mAP: +3.7 mAP | Improved |
| use smallest box instead of closest box for undecided points (ablation) | Box2Mask · ScanNet val | mAP@50 | Improved |
| use Box2Mask voting and clustering instead of detector + box-to-point segmentation (ablation) | Box2Mask · ScanNet val | mAP@50: 47.9 -> 59.7 | Improved |
| use Box2Mask voting and clustering instead of detector + box-to-point segmentation (ablation) | Box2Mask · ScanNet val | mAP: 26.5 -> 39.1 | Improved |
| vote bounding boxes with NMC instead of centers with spatial clustering (ablation) | Box2Mask · ScanNet val | mAP: +8 mAP | Improved |
| condition clustering on semantic class (ablation) | center votes vs box votes · ScanNet val | mAP | Mixed |
| use NMC instead of spatial clustering (ablation) | Box2Mask · ScanNet val | mAP | Improved |
| add noisy box labels (0 to 20 cm corner error, up to 10% missing) (ablation) | Box2Mask · ScanNet val | mAP: around 4 mAP | No clear effect |
| set NMC clustering threshold tau to 0.3 (ablation) | Box2Mask · ScanNet val | mAP | Improved |
| train with dense per-point labels instead of box labels (ablation) | Box2Mask · ScanNet val | mAP@50: 59.7 -> 64.7 | Improved |
| train with bounding boxes instead of 20 annotated points | Box2Mask vs CSC · ScanNet data efficient | mAP@50: 28.9 -> 46.5 | Improved |
Closest papers by meaning
- Sketchy Bounding-box Supervision for 3D Instance Segmentation (CVPR 2025 · similarity 0.93)
- Simple Does It: Weakly Supervised Instance and Semantic Segmentation (CVPR 2017 · similarity 0.91)
- DBGroup: Dual-Branch Point Grouping for Weakly Supervised 3D Semantic Instance Segmentation (AAAI 2026 · similarity 0.91)
- Weakly Supervised Semantic Segmentation via Box-Driven Masking and Filling Rate Shifting (TPAMI 2023 · similarity 0.90)
- Box2Seg: Attention Weighted Loss and Discriminative Feature Learning for Weakly Supervised Segmentation (ECCV 2020 · similarity 0.90)
- BSNet: Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation (CVPR 2024 · similarity 0.88)
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.
