Research map / Point clouds / Segmentation and detection
SAM-guided Graph Cut for 3D Instance Segmentation
Haoyu Guo, He Zhu, Sida Peng, Yuang Wang, Yujun Shen, Ruizhen Hu, Xiaowei Zhou
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 GNN refinement on top of SAM-annotated graph (ablation) | SAM-guided graph cut · ScanNet200 | mAP: 19.7 -> 22.1 | Improved |
| add SAM-derived edge weights to the graph (ablation) | SAM-guided graph cut · ScanNet200 | mAP: 10.1 -> 22.1 | Improved |
| add SAM node features (ablation) | SAM-guided graph cut · ScanNet200 | mAP: 19.4 -> 22.1 | Improved |
| add regularization loss for edges outside pseudo-labels (ablation) | GNN training · ScanNet200 | mAP: 19.9 -> 22.1 | Improved |
| increase prompt points k from 1 to 5 (ablation) | SAM-guided graph cut · ScanNet200 | mAP: 20.3 -> 22.1 | Improved |
| increase prompt points k from 5 to 9 (ablation) | SAM-guided graph cut · ScanNet200 | mAP: 22.1 -> 21.9 | Hurt |
| increase min co-visible views n from 10 to 30 (ablation) | pseudo-label generation · ScanNet200 | mAP: 22.1 -> 20.2 | Hurt |
| use CropFormer instead of SAM to generate pseudo-labels (ablation) | GNN training · ScanNet200 pseudo-label quality | accuracy: 0.892 -> 0.912 | Improved |
| use direct graph partition (union-find) instead of NCuts or DBSCAN (ablation) | SAM-annotated graph · ScanNet200 | mAP: 15.7 -> 19.7 (NCuts); 10.3 -> 19.7 (DBSCAN) | Improved |
| train GNN on ScanNet200 pseudo-labels, test on other datasets (ablation) | SAM-guided graph cut · KITTI-360 | mAP: 22.6 -> 23.8 | Improved |
| use supervised Mask3D instead of the pseudo-label method on out-of-domain data | Mask3D · KITTI-360 | mAP: 23.8 -> 0.2 | Hurt |
| add GNN to NCuts graph cut on KITTI-360 (ablation) | SAM graph + NCuts · KITTI-360 | mAP: 19.5 -> 18.1 | Hurt |
Closest papers by meaning
- SA3DIP: Segment Any 3D Instance with Potential 3D Priors (NeurIPS 2024 · similarity 0.90)
- SAM2Object: Consolidating View Consistency via SAM2 for Zero-Shot 3D Instance Segmentation (CVPR 2025 · similarity 0.89)
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · similarity 0.89)
- Point-SAM: Promptable 3D Segmentation Model for Point Clouds (ICLR 2025 · similarity 0.89)
- SegGraph: Leveraging Graphs of SAM Segments for Few-Shot 3D Part Segmentation (NeurIPS 2025 · similarity 0.88)
- Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025 · 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.
