Research map / Point clouds / Segmentation and detection
Scalable SoftGroup for 3D Instance Segmentation on Point Clouds
Thang Vu, Kookhoi Kim, Thanh Nguyen, Tung M. Luu, Junyeong Kim, Chang D. Yoo
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 soft grouping (ablation) | hard-grouping baseline · ScanNet v2 val | AP50: 61.1 -> 63.8 | Improved |
| add top-down refinement (ablation) | hard-grouping baseline · ScanNet v2 val | AP50: 61.1 -> 65.4 | Improved |
| add top-down refinement to soft grouping (ablation) | SoftGroup · ScanNet v2 val | AP50: 63.8 -> 67.6 | Improved |
| use soft grouping score threshold 0.01 instead of hard prediction (ablation) | SoftGroup · ScanNet v2 val | AP50: 65.4 -> 58.5 | Hurt |
| set soft grouping threshold 0.2 instead of none (ablation) | SoftGroup · ScanNet v2 val | AP: 44.3 -> 46.0 | Improved |
| add mask branch to class-only top-down refinement (ablation) | SoftGroup · ScanNet v2 val | AP: 41.1 -> 45.7 | Improved |
| add mask scoring branch (ablation) | SoftGroup · ScanNet v2 val | AP50: 68.4 -> 67.6 | Mixed |
| replace naive scaling with class-aware pyramid scaling (CAPS) (ablation) | SoftGroup++ · S3DIS Area 5 | AP: 47.0 -> 50.9 | Improved |
| add octree k-NN (ablation) | SoftGroup++ baseline · S3DIS largest scene | total runtime (s): 30.32 -> 9.85 | Improved |
| add late devoxelization (ablation) | SoftGroup++ baseline · S3DIS largest scene | total runtime (s): 30.32 -> 2.05 | Improved |
| add octree k-NN to CAPS plus late devoxelization (ablation) | SoftGroup++ · S3DIS largest scene | k-NN runtime (s): 0.50 -> 0.44 | Improved |
| add octree k-NN (ablation) | SoftGroup++ · STPLS3D uncropped | k-NN runtime (s): 1.48 -> 0.93 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PointGroup — PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation (CVPR 2020) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SoftGroup for 3D Instance Segmentation on Point Clouds (CVPR 2022 · similarity 0.94)
- PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation (CVPR 2020 · similarity 0.90)
- DBGroup: Dual-Branch Point Grouping for Weakly Supervised 3D Semantic Instance Segmentation (AAAI 2026 · similarity 0.88)
- SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation (CVPR 2018 · similarity 0.87)
- SPGroup3D: Superpoint Grouping Network for Indoor 3D Object Detection (AAAI 2024 · similarity 0.86)
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.85)
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.
