Research map / 3D vision / 3D generation
SpaCeFormer: Fast Proposal-Free Open-Vocabulary 3D Instance Segmentation
Chris Choy, Junha Lee, Chunghyun Park, Minsu Cho, Jan Kautz
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 Morton curve attention to window attention (ablation) | SpaCeFormer backbone · ScanNet200 val | mAP: 0.09517 -> 0.11092 | Improved |
| add window attention to Morton-only attention (ablation) | SpaCeFormer backbone · ScanNet200 val | mAP: 0.09470 -> 0.11092 | Improved |
| add Morton attention to window-only on class-agnostic AP (ablation) | SpaCeFormer backbone · ScanNet200 val | AP: 0.25466 -> 0.25178 | No clear effect |
| use RoPE in decoder instead of no positional encoding (ablation) | SpaCeFormer decoder · ScanNet200 val | mAP: 5.97 -> 7.60 | Improved |
| use RoPE instead of absolute positional embedding (ablation) | SpaCeFormer decoder · ScanNet200 val | mAP: 5.95 -> 7.60 | Improved |
| use RoPE instead of learnable positional bias (ablation) | SpaCeFormer decoder · ScanNet200 val | mAP: 6.46 -> 7.60 | Improved |
| use learned queries instead of random init (ablation) | SpaCeFormer decoder · ScanNet200 val | class-agnostic AP: 0.14635 -> 0.19819 | Improved |
| use learned queries instead of FPS init (ablation) | SpaCeFormer decoder · ScanNet200 val | class-agnostic AP: 0.17970 -> 0.19819 | Improved |
| use learned queries instead of FPS init on class-wise mAP (ablation) | SpaCeFormer decoder · ScanNet200 val | mAP: 0.07164 -> 0.06443 | Hurt |
| multi-view mask clustering instead of naive SAM2 2D-to-3D lifting | SpaCeFormer-3M masks · 312 ScanNet train scenes | recall@0.5: 2.5% -> 54.3% | Improved |
| replace Mask3D proposals with Segment3D open-vocab proposals | Mosaic3D · ScanNet200 | mAP: 11.8 -> 2.7 | Hurt |
| replace Segment3D proposals with proposal-free decoder | Mosaic3D / SpaCeFormer · ScanNet200 | mAP: 2.7 -> 11.1 | Improved |
Closest papers by meaning
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.90)
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · similarity 0.88)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.88)
- Open-YOLO 3D: Towards Fast and Accurate Open-Vocabulary 3D Instance Segmentation (ICLR 2025 · similarity 0.88)
- Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation (CVPR 2025 · similarity 0.88)
- MaskClustering: View Consensus based Mask Graph Clustering for Open-Vocabulary 3D Instance Segmentation (CVPR 2024 · similarity 0.87)
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.
