Research map / Detection and segmentation / Open-set and few-shot detection
OpenMask3D: Open-Vocabulary 3D Instance Segmentation
Ayça Takmaz, Elisabetta Fedele, Robert Sumner, Marc Pollefeys, Federico Tombari, Francis Engelmann
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 SAM-refined 2D masks for cropping (ablation) | OpenMask3D · ScanNet200 val | AP: 12.9 -> 14.1 | Improved |
| add multi-scale crops (3 levels) (ablation) | OpenMask3D · ScanNet200 val | AP: 12.9 -> 14.3 | Improved |
| combine SAM masks with multi-scale crops (ablation) | OpenMask3D · ScanNet200 val | AP: 12.9 -> 15.4 | Improved |
| increase top-k views from 1 to 5 (ablation) | OpenMask3D · ScanNet200 val | AP: 13.6 -> 15.4 | Improved |
| increase top-k views from 5 to 10 (ablation) | OpenMask3D · ScanNet200 val | AP: 15.4 -> 15.4 | No clear effect |
| use 3 crop levels instead of 1 (ablation) | OpenMask3D · Replica | AP: 11.3 -> 13.1 | Improved |
| use 5 crop levels instead of 3 (ablation) | OpenMask3D · Replica | AP: 13.1 -> 12.8 | Hurt |
| expansion ratio 0.05 or 0.2 instead of 0.1 (ablation) | OpenMask3D · Replica | AP: 13.1 -> 12.9 / 12.8 | Hurt |
| use ground truth oracle masks instead of predicted masks | OpenMask3D · ScanNet200 val | AP: 15.4 -> 29.1 | Improved |
| replace real RGB-D images with images rendered from the point cloud | OpenMask3D · Replica | AP: 13.1 -> 11.6 | Hurt |
| instance-mask CLIP features instead of OpenScene point features averaged over masks | OpenMask3D · ScanNet200 val | AP: 11.7 -> 15.4 | Improved |
| train mask module on ScanNet20 instead of ScanNet200 | OpenMask3D · ScanNet200 val | AP: 15.4 -> 12.6 | Hurt |
Later papers that compared with this method
7 later accepted papers report OpenMask3D as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- Generalized Decoupled Learning for Enhancing Open-Vocabulary Dense Perception (TPAMI 2026 · better on 1 of 1 shared result)
- Segment Any 3D Object with Language (ICLR 2025 · better on 4 of 5 shared results)
- Find Any Part in 3D (ICCV 2025 · better on 5 of 5 shared results)
- OV3D-CG: Open-vocabulary 3D Instance Segmentation with Contextual Guidance (ICCV 2025 · better on 2 of 2 shared results)
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · better on 7 of 9 shared results)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · better on 2 of 2 shared results)
- Open-Vocabulary 3D Semantic Segmentation with Text-to-Image Diffusion Models (ECCV 2024 · better on 1 of 1 shared result)
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
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.95)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.94)
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · similarity 0.93)
- Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation (CVPR 2025 · similarity 0.92)
- Open-YOLO 3D: Towards Fast and Accurate Open-Vocabulary 3D Instance Segmentation (ICLR 2025 · similarity 0.92)
- Open-Vocabulary 3D Semantic Segmentation with Foundation Models (CVPR 2024 · similarity 0.92)
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.
