Research map / Point clouds / Segmentation and detection
Part2Object: Hierarchical Unsupervised 3D Instance Segmentation
Cheng Shi, Yulin Zhang, Bin Yang, Jiajin Tang, Yuexin Ma, Sibei Yang
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 |
|---|---|---|---|
| use hierarchical multi-layer clustering instead of single-layer clustering (ablation) | Part2Object · ScanNet val | mAP@50: 13.8 -> 26.8 | Improved |
| add 3D objectness guidance (stop criterion) (ablation) | Part2Object · ScanNet val | mAP@50: 20.7 -> 26.8 | Improved |
| use similarity-weighted cluster features instead of plain average (ablation) | Part2Object · ScanNet val | mAP: 10.6 -> 12.6 | Improved |
| add object-part queries (Hi-Mask3D) over Mask3D (ablation) | Mask3D on Part2Object pseudo-labels · ScanNet val | mAP@50: 31.0 -> 36.0 | Improved |
| add rounds of self-training (ablation) | Hi-Mask3D · ScanNet val | mAP: 15.3 -> 16.9 | Improved |
| vary merge ratio K (0.4 to 0.7) (ablation) | Part2Object · ScanNet val | mAP@50: K=0.6 gives 26.8, K=0.7 gives 21.1, K=0.4 gives 22.6 | Mixed |
| use Part2Object pseudo-label pre-training instead of unsupervised baselines Unscene3D | Hi-Mask3D · ScanNet val (1% labels) | mAP@50: 28.4 -> 44.1 | Improved |
| train on unsupervised Part2Object pseudo-labels instead of fully supervised ScanNet labels | Hi-Mask3D · ScanNet200 zero-shot | mAP@50: 24.2 -> 34.9 | Improved |
| pre-train on Part2Object pseudo-labels before data-efficient fine-tuning | Hi-Mask3D · ScanNet200 (1% data) | mAP@50: 1.3 -> 35.4 | Improved |
Closest papers by meaning
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · similarity 0.90)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.90)
- UnScene3D: Unsupervised 3D Instance Segmentation for Indoor Scenes (CVPR 2024 · similarity 0.89)
- MaskClustering: View Consensus based Mask Graph Clustering for Open-Vocabulary 3D Instance Segmentation (CVPR 2024 · similarity 0.89)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.89)
- Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation (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.
