Research map / Detection and segmentation / Open-set and few-shot detection
GeoGuide: Hierarchical Geometric Guidance for Open-Vocabulary 3D Semantic Segmentation
Xujing Tao, Chuxin Wang, Yubo Ai, Zhixin Cheng, Zhuoyuan Li, Liangsheng Liu, Yujia Chen, Xinjun Li, Qiao Li, Wenfei Yang, Tianzhu Zhang
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 |
|---|---|---|---|
| replace average-pooling superpoint distillation with uncertainty-weighted superpoint distillation (USD) (ablation) | GeoGuide on LSeg · ScanNet v2 val | mIoU: 55.8 -> 56.4 | Improved |
| add instance-level mask reconstruction (IMR) alone (ablation) | GeoGuide on LSeg · ScanNet v2 val | mIoU: 53.9 -> 56.8 | Improved |
| add inter-instance relation consistency (IIRC) alone (ablation) | GeoGuide on LSeg · ScanNet v2 val | mIoU: 53.9 -> 57.1 | Improved |
| add IMR on top of USD (ablation) | GeoGuide on LSeg · ScanNet v2 val | mIoU: 56.4 -> 57.9 | Improved |
| add IIRC on top of USD and IMR (ablation) | GeoGuide on LSeg · ScanNet v2 val | mIoU: 57.9 -> 59.8 | Improved |
| swap in a pretrained 3D backbone without geometric constraints (ablation) | OpenScene baseline with LSeg · ScanNet v2 val | mIoU: 54.2 -> 53.9 | No clear effect |
| load pretrained weights vs training the same backbone from scratch (ablation) | 3D backbone with LSeg distillation · ScanNet v2 val | mIoU: 53.1 -> 53.9 | Improved |
| freeze pretrained 3D backbone and train only an MLP adapter (OpenSeg) (ablation) | OpenScene distillation · ScanNet v2 | mIoU: 47.5 -> 50.4 | Improved |
| freeze pretrained 3D backbone and train only an MLP adapter (LSeg) (ablation) | OpenScene distillation · ScanNet v2 | mIoU: 54.2 -> 53.9 | No clear effect |
| use GeoGuide with SAS* features vs SAS stage1 baseline | GeoGuide · ScanNet v2, nuScenes, Matterport3D | mIoU: +5.6, +4.9, +5.6 mIoU | Improved |
| use GeoGuide with LSeg features on nuScenes | GeoGuide · nuScenes val | mIoU: 36.7 -> 40.4 | Mixed |
| GeoGuide (SAS* features) at large category counts on Matterport3D | GeoGuide · Matterport3D K=40 | mIoU: 33.8 vs OpenSeg variant 38.5 | Mixed |
Closest papers by meaning
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.90)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.89)
- Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation (NeurIPS 2026 · similarity 0.89)
- XMask3D: Cross-modal Mask Reasoning for Open Vocabulary 3D Semantic Segmentation (NeurIPS 2024 · similarity 0.89)
- Open-Vocabulary 3D Semantic Segmentation with Foundation Models (CVPR 2024 · similarity 0.89)
- Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation (CVPR 2025 · 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.
