Research map / Point clouds / Segmentation and detection
See More and Know More: Zero-shot Point Cloud Segmentation via Multi-modal Visual Data
Yuhang Lu, Qi Jiang, Runnan Chen, Yuenan Hou, Xinge Zhu, Yuexin Ma
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 semantic-guided visual feature fusion (SGVF) instead of concatenation (ablation) | multi-modal zero-shot segmentation · SemanticKITTI (4 unseen) | unseen mIoU: 23.4 -> 26.8 | Improved |
| add semantic-visual feature enhancement (SVFE) (ablation) | multi-modal zero-shot segmentation · SemanticKITTI (4 unseen) | unseen mIoU: 19.9 -> 26.8 | Improved |
| add image modality to LiDAR (ablation) | multi-modal zero-shot segmentation · SemanticKITTI (4 unseen) | unseen mIoU: 20.0 -> 26.8 | Improved |
| use SVFE cross-attention instead of equal-size self-attention (ablation) | baseline TGP · SemanticKITTI (4 unseen) | unseen mIoU: 19.4 -> 23.4 | Improved |
| enhance semantic features with point features before image features (ablation) | SVFE · SemanticKITTI (4 unseen) | unseen mIoU: 16.1 -> 26.8 | Improved |
| add self-attention on top of SGVF fused features (ablation) | SGVF · SemanticKITTI (4 unseen) | unseen mIoU: 26.8 -> 21.2 | Hurt |
| use CLIP text and MaskCLIP image features (ablation) | multi-modal zero-shot segmentation · SemanticKITTI (4 unseen) | unseen mIoU: 26.8 -> 14.1 | Hurt |
| use existing camera-LiDAR fusion methods with TGP baseline | TGP baseline · SemanticKITTI (4 unseen) | unseen mIoU: TGP 17.3; fusion methods 11.9 to 16.3 | Mixed |
| switch image encoder from ResUnet-34 to ResUnet-50 or ResUnet-18 (ablation) | multi-modal zero-shot segmentation · SemanticKITTI (4 unseen) | unseen mIoU: ResUnet-18 24.7, ResUnet-34 26.8, ResUnet-50 27.1 | No clear effect |
| use multi-modal method at 8 unseen classes | multi-modal zero-shot segmentation · SemanticKITTI (8 unseen) | seen mIoU: 53.2 -> 46.0 | Hurt |
Closest papers by meaning
- Zero-Shot Point Cloud Segmentation by Semantic-Visual Aware Synthesis (ICCV 2023 · similarity 0.92)
- Generalized Zero-Shot Learning for Point Cloud Segmentation with Evidence-Based Dynamic Calibration (AAAI 2025 · similarity 0.89)
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.88)
- Multimodality Helps Few-shot 3D Point Cloud Semantic Segmentation (ICLR 2025 · similarity 0.88)
- Generalized Few-shot 3D Point Cloud Segmentation with Vision-Language Model (CVPR 2025 · similarity 0.87)
- Taylor Series-Inspired Local Structure Fitting Network for Few-shot Point Cloud Semantic Segmentation (AAAI 2025 · 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.
