Research map / Point clouds / Segmentation and detection
Position-Guided Point Cloud Panoptic Segmentation Transformer
Zeqi Xiao, Wenwei Zhang, Tai Wang, Chen Change Loy, Dahua Lin, Jiangmiao Pang
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 Mixed-parameterized Positional Embedding (MPE) (ablation) | DETR-style baseline · SemanticKITTI val | PQ: 58.0 -> 59.8 | Improved |
| add MPE (ablation) | baseline · SemanticKITTI val | PQ Things: 59.6 -> 62.4 | Improved |
| add Position-Aware Segmentation (PA-Seg) (ablation) | baseline + MPE · SemanticKITTI val | PQ: 59.8 -> 60.9 | Improved |
| add PA-Seg (ablation) | baseline + MPE · SemanticKITTI val | PQ Things: 62.4 -> 65.1 | Improved |
| add Masked Focal Attention (MFA) (ablation) | baseline + MPE + PA-Seg · SemanticKITTI val | PQ: 60.9 -> 61.4 | Improved |
| add MFA (ablation) | baseline + MPE + PA-Seg · SemanticKITTI val | PQ Stuff: 57.9 -> 57.7 | No clear effect |
| add PA-Seg at refinement layer 1 (ablation) | P3Former · SemanticKITTI val | PQ: 58.3 -> 59.8 | Improved |
| add PA-Seg at layer 0 (before refinement) (ablation) | P3Former · SemanticKITTI val | PQ: 55.8 -> 56.0 | No clear effect |
| use voxel features instead of point features (ablation) | P3Former · SemanticKITTI val | PQ: 59.4 -> 60.9 | Improved |
| use focal loss instead of cross-entropy (ablation) | P3Former · SemanticKITTI val | PQ: 58.9 -> 59.0 | No clear effect |
| raise dice loss weight from 1 to 2 (ablation) | P3Former · SemanticKITTI val | PQ / SQ: PQ 59.0 -> 59.8, SQ 75.1 -> 72.7 | Mixed |
| apply P3Former head to semantic segmentation instead of Cylinder3D decode head | Cylinder3D · SemanticKITTI | mIoU: 62.5 -> 64.2 | Improved |
Closest papers by meaning
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.89)
- Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR-Based Perception (TPAMI 2021 · similarity 0.89)
- Panoptic-PolarNet: Proposal-Free LiDAR Point Cloud Panoptic Segmentation (CVPR 2021 · similarity 0.88)
- LiDAR-Based Panoptic Segmentation via Dynamic Shifting Network (CVPR 2021 · similarity 0.88)
- How Do Images Align and Complement LiDAR? Towards a Harmonized Multi-modal 3D Panoptic Segmentation (ICML 2025 · similarity 0.88)
- Beyond the Label Itself: Latent Labels Enhance Semi-supervised Point Cloud Panoptic Segmentation (AAAI 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.
