Research map / Detection and segmentation / Open-set and few-shot detection
3D Weakly Supervised Semantic Segmentation with 2D Vision-Language Guidance
Xiaoxu Xu, Yitian Yuan, Jinlong Li, Qiudan Zhang, Zequn Jie, Lin Ma, Hao Tang, Nicu Sebe, Xu Wang
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 scene-level label filtering to pseudo labels (ablation) | baseline MinkowskiNet18A · S3DIS | mIoU: 37.7 -> 42.6 | Improved |
| add Embeddings Soft-Guidance Stage instead of direct pseudo-label supervision (ablation) | baseline MinkowskiNet18A · S3DIS | mIoU: 37.7 -> 38.2 | Improved |
| add filtering and Embeddings Specialization Stage on top of soft-guidance (ablation) | 3DSS-VLG · S3DIS | mIoU: 38.2 -> 45.3 | Improved |
| use text-embedding guidance from 2D vision-language model instead of CAM | 3D WSSS with scene-level labels · S3DIS | mIoU: 27.7 -> 45.3 versus MIT | Improved |
| 3DSS-VLG with scene-level labels versus Kweon et al. with extra image-level labels | 3DSS-VLG · ScanNet test | mIoU: 47.4 -> 48.9 | Improved |
| change 3D backbone from MinkowskiNet18A to 14A or 34A (ablation) | 3DSS-VLG · S3DIS | mIoU: 45.3 -> 44.5 (14A), 44.7 (34A) | Hurt |
| use Transformers instead of MLP in ESS adapter (ablation) | 3DSS-VLG · S3DIS | mIoU: 45.3 -> 45.0 | No clear effect |
| train on S3DIS and test on ScanNet versus the reverse | 3DSS-VLG · cross-dataset generalization | mIoU: S3DIS->ScanNet 13.4; ScanNet->S3DIS 33.3 | Mixed |
Closest papers by meaning
- Joint Learning of 2D-3D Weakly Supervised Semantic Segmentation (NeurIPS 2022 · similarity 0.90)
- Rewis3d: Reconstruction Improves Weakly-Supervised Semantic Segmentation (CVPR 2026 · similarity 0.89)
- Box2Mask: Weakly Supervised 3D Semantic Instance Segmentation Using Bounding Boxes (ECCV 2022 · similarity 0.88)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.87)
- econSG: Efficient and Multi-view Consistent Open-Vocabulary 3D Semantic Gaussians (ICLR 2025 · similarity 0.87)
- LESS: Label-Efficient and Single-Stage Referring 3D Segmentation (NeurIPS 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.
