Research map / Point clouds / Segmentation and detection
Zero-Shot Point Cloud Segmentation by Semantic-Visual Aware Synthesis
Yuwei Yang, Munawar Hayat, Zhao Jin, Hongyuan Zhu, Yinjie Lei
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 Mask Correspondence Learning (masking + contrastive) to the generator (ablation) | 3DGenZ-style baseline · ScanNet | HmIoU: 12.5 -> 18.8 | Improved |
| add Heterogeneous Prototype Alignment (ablation) | baseline · ScanNet | HmIoU: 12.5 -> 15.0 | Improved |
| add Relational Transfer Consistency (ablation) | baseline · ScanNet | HmIoU: 12.5 -> 15.4 | Improved |
| add all three modules (MCL, HPA, RTC) (ablation) | baseline · ScanNet | HmIoU: 12.5 -> 20.2 | Improved |
| add contrastive learning without masking (ablation) | baseline · ScanNet | seen mIoU: 34.2 -> 31.7 | Hurt |
| add masking strategy on top of contrastive learning (ablation) | MCL · ScanNet | HmIoU: 16.8 -> 18.8 | Improved |
| use neighbor-aware prototypes instead of simple averaging in HPA (ablation) | HPA · ScanNet | HmIoU: 19.6 -> 20.2 | Improved |
| use neighbor-aware prototypes instead of simple averaging in RTC (ablation) | RTC · ScanNet | HmIoU: 20.2 -> 17.7 | Hurt |
| use K-Means prototypes instead of simple averaging in RTC (ablation) | RTC · ScanNet | HmIoU: 20.2 -> 18.9 | Hurt |
| increase mask probability q beyond 0.2 (ablation) | MCL · ScanNet, S3DIS, SemanticKITTI | HmIoU | Hurt |
| increase visual prototype ratio r beyond 0.04 (ablation) | HPA · ScanNet, S3DIS, SemanticKITTI | HmIoU | Hurt |
| use GloVe+Word2Vec instead of GloVe alone on SemanticKITTI (ablation) | Ours · SemanticKITTI | HM: 10.8 -> 20.1 | Improved |
Closest papers by meaning
- Generalized Zero-Shot Learning for Point Cloud Segmentation with Evidence-Based Dynamic Calibration (AAAI 2025 · similarity 0.90)
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.90)
- Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.89)
- LogoSP: Local-global Grouping of Superpoints for Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2025 · similarity 0.89)
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.89)
- Generalized Few-shot 3D Point Cloud Segmentation with Vision-Language Model (CVPR 2025 · similarity 0.88)
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.
