Research map / Detection and segmentation / Open-set and few-shot detection
Details Matter for Indoor Open-vocabulary 3D Instance Segmentation
Sanghun Jung, Jingjing Zheng, Ke Zhang, Nan Qiao, Albert Y. C. Chen, Lu Xia, Chi Liu, Yuyin Sun, Xiao Zeng, Hsiang-Wei Huang, Byron Boots, Min Sun and 1 more
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 |
|---|---|---|---|
| use frame-wise sIOU instead of tracklet-wise sIOU for tracking (ablation) | image-based proposals · ScanNet200 val subset class-agnostic | AP50: 54.3 -> 56.1 | Improved |
| add iterative merging/removal (ablation) | aggregation only · ScanNet200 val subset class-agnostic | AP25: 63.5 -> 68.5 | Improved |
| add overlap removal in 2D grounding (ablation) | aggregation + merging/removal · ScanNet200 val subset class-agnostic | AP: 31.8 -> 33.5 | Improved |
| add iterative refinement with multi-view consensus (ablation) | aggregation + merging + overlap removal · ScanNet200 val subset class-agnostic | AP: 33.5 -> 35.1 | Improved |
| replace CLIP with Alpha-CLIP (ablation) | OV-3DIS pipeline · ScanNet200 val Top-K 2D+3D | mAP: 27.5 -> 30.5 | Improved |
| add SMS-based filtering (ablation) | OV-3DIS pipeline with Alpha-CLIP · ScanNet200 val Top-K 2D+3D | mAP: 30.5 -> 32.7 | Improved |
| combine image-based and point-cloud-based proposals | Ours · ScanNet200 val Top-K | mAP: 25.4 -> 32.7 | Improved |
| combine image-based and point-cloud-based proposals | Ours · S3DIS | mAR: 35.3 -> 48.2 | Improved |
| use Alpha-CLIP classification (vs OpenYOLO3D) with ScanNet200-trained proposals | Ours (3D only) · Replica Top-1 | mAP: 23.7 -> 22.0 | Hurt |
| iterative merging/removal on small-object dataset | Ours (2D only) · ScanNet++ class-agnostic | AP: 18.5 -> 18.2 | No clear effect |
| vary image visibility threshold tau_img from 0.1 (ablation) | image-based proposals · ScanNet200 val subset class-agnostic | AP: 35.1 -> 16.2 | Hurt |
| vary merge threshold tau_merge from 0.3 to 0.1 (ablation) | image-based proposals · ScanNet200 val subset class-agnostic | AP: 35.1 -> 36.1 | Improved |
Closest papers by meaning
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.91)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.89)
- Open-YOLO 3D: Towards Fast and Accurate Open-Vocabulary 3D Instance Segmentation (ICLR 2025 · similarity 0.89)
- Open-Vocabulary 3D Semantic Segmentation with Foundation Models (CVPR 2024 · similarity 0.88)
- COS3D: Collaborative Open-Vocabulary 3D Segmentation (NeurIPS 2025 · similarity 0.88)
- SAI3D: Segment Any Instance in 3D Scenes (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.
