Research map / Point clouds / Segmentation and detection
SoftGroup for 3D Instance Segmentation on Point Clouds
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace hard grouping with soft semantic-score grouping (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 39.5 -> 41.6 | Helped |
| add top-down refinement to hard grouping (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 39.5 -> 44.3 | Helped |
| combine soft grouping and top-down refinement (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: +6.5 | Helped |
| set soft-grouping threshold to 0.2 instead of hard semantic grouping (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 44.3 -> 46.0 | Helped |
| set soft-grouping threshold to 0.01 (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 44.3 -> 40.1 | Hurt |
| vary the soft-grouping threshold to 0.1, 0.3, or 0.5 rather than 0.2 (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 45.3; 45.2; 43.9 vs 46.0 | Hurt |
| add the mask branch to the classification branch (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 41.1 -> 45.7 | Helped |
| add the mask-scoring branch to class and mask branches (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 45.7 -> 46.0 | Helped |
| derive instance category from the classification branch instead of semantic majority vote (ablation) | SoftGroup on ScanNet v2 validation · ScanNet v2 validation | AP: 45.0 -> 46.0 | Helped |
| use SoftGroup rather than SSTNet | 3D instance segmentation · ScanNet v2 hidden test | AP50: 69.8 -> 76.1 | Helped |
| use SoftGroup rather than SSTNet | 3D instance segmentation · S3DIS Area 5 | AP50: 59.3 -> 66.1 | Helped |
| use SoftGroup rather than SSTNet | 3D instance segmentation · S3DIS 6-fold cross-validation | AP50: 67.8 -> 68.9 | Helped |
Closest papers by meaning
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.84)
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · similarity 0.83)
- Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation (CVPR 2024 · similarity 0.83)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.83)
- Insightful Instance Features for 3D Instance Segmentation (CVPR 2025 · similarity 0.83)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.82)
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.