Research map / Detection and segmentation / 3D detection and driving
Sparse Cross-Scale Attention Network for Efficient LiDAR Panoptic Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use SCAN instead of GP-S3Net | LiDAR panoptic segmentation · SemanticKITTI test | PQ: 60.0 -> 61.5 | Helped |
| Use SCAN instead of EfficientLPS | LiDAR panoptic segmentation · nuScenes validation | PQ: 59.2 -> 65.1 | Helped |
| Add multi-scale sparse supervision (ablation) | SCAN backbone with dense centroid distribution · SemanticKITTI validation | PQ: 53.3 -> 53.7 | Helped |
| Add cross-scale global attention (ablation) | SCAN backbone with dense centroid distribution · SemanticKITTI validation | PQ: 53.3 -> 56.0 | Helped |
| Add multi-scale sparse supervision alongside cross-scale global attention (ablation) | SCAN with dense centroid distribution · SemanticKITTI validation | PQ: 56.0 -> 56.7 | Helped |
| Replace dense centroid distribution with 3D sparse centroid distribution (ablation) | SCAN with MSS and CGA · SemanticKITTI validation | PQ: 56.7 -> 56.5 | Hurt |
| Replace dense centroid distribution with BEV sparse centroid distribution (ablation) | SCAN with MSS and CGA · SemanticKITTI validation | PQ: 56.7 -> 57.2 | Helped |
| Replace dense centroid distribution with BEV sparse centroid distribution (ablation) | SCAN with MSS and CGA · SemanticKITTI validation | Runtime: 81ms -> 78ms | Helped |
| Use b3-b4 cross-attention instead of b4-only self-attention (ablation) | Cross-scale global attention · SemanticKITTI validation | PQ: 54.5 -> 55.9 | Helped |
| Include b2 features with b3-b4 in attention (ablation) | Cross-scale global attention · SemanticKITTI validation | PQ: 55.9 -> 56.8 | Helped |
| Include b1 features in attention (ablation) | Cross-scale global attention · SemanticKITTI validation | PQ: 56.8 -> 56.6 | Hurt |
| Remove 3D position encoding (ablation) | Cross-scale global attention · SemanticKITTI validation | PQ: 57.2 -> 55.6 | Hurt |
Closest papers by meaning
- Panoptic-PHNet: Towards Real-Time and High-Precision LiDAR Panoptic Segmentation via Clustering Pseudo Heatmap (CVPR 2022 · similarity 0.90)
- Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR Segmentation (CVPR 2021 · similarity 0.89)
- How Do Images Align and Complement LiDAR? Towards a Harmonized Multi-modal 3D Panoptic Segmentation (ICML 2025 · similarity 0.88)
- LSK3DNet: Towards Effective and Efficient 3D Perception with Large Sparse Kernels (CVPR 2024 · similarity 0.88)
- PC-BEV: An Efficient Polar-Cartesian BEV Fusion Framework for LiDAR Semantic Segmentation (AAAI 2025 · similarity 0.87)
- 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.