Research map / Point clouds / Segmentation and detection
CutS3D: Cutting Semantics in 3D for 2D Unsupervised Instance Segmentation
Leon Sick, Dominik Engel, Sebastian Hartwig, Pedro Hermosilla, Timo Ropinski
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 LocalCut (ablation) | CutLER on DiffNCuts · COCO val2017 (1 round) | AP mask: 9.4 -> 9.5 | Improved |
| add Spatial Importance sharpening (ablation) | CutLER + LocalCut · COCO val2017 (1 round) | AP mask: 9.5 -> 9.8 | Improved |
| add Spatial Confidence (ablation) | CutLER + LocalCut + Spatial Importance · COCO val2017 (1 round) | AP mask: 9.8 -> 10.2 | Improved |
| add 3D cutting contributions overall (ablation) | CutLER · COCO val2017 (1 round) | AP box 50: 22.1 -> 23.9 | Improved |
| add confident copy-paste (ablation) | CAD without Spatial Confidence · COCO val2017 (no self-training) | AP mask: 8.5 -> 8.8 | Improved |
| add alpha blending with Spatial Confidence (ablation) | CAD with copy-paste · COCO val2017 (no self-training) | AP mask: 8.8 -> 9.0 | Improved |
| add Spatial Confidence soft loss (ablation) | CAD with copy-paste and alpha blend · COCO val2017 (no self-training) | AP mask: 9.0 -> 9.1 | Improved |
| use spatial confidence maps instead of scalar average confidence for alpha (ablation) | CutS3D · COCO val2017 | AP mask: 8.8 -> 9.0 | Improved |
| use ZoeDepth instead of MiDaS (Small) depth (ablation) | CutS3D · COCO val2017 | AP mask: 8.7 -> 9.1 | Improved |
| use DiffNCuts backbone instead of DINO for CutLER (ablation) | CutLER · COCO val2017 (round 3) | AP mask: 9.7 -> 9.8 | No clear effect |
| compare CutS3D with CutLER at equal self-training rounds | CutLER (DINO) · COCO val2017 | AP mask: 8.8 -> 10.0 | Improved |
| use CutS3D instead of CuVLER | CuVLER · Pascal VOC (zero-shot detection) | AP box: 22.3 -> 21.4 | Hurt |
Closest papers by meaning
- 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 0.90)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.89)
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · similarity 0.89)
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · similarity 0.89)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.89)
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.
