Research map / Detection and segmentation / 3D detection and driving
Beyond One Shot, Beyond One Perspective: Cross-View and Long-Horizon Distillation for Better LiDAR Representations
Xiang Xu, Lingdong Kong, Song Wang, Chuanwei Zhou, Qingshan Liu
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 cross-view aggregation (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 51.90 -> 53.72 | Improved |
| add long-term feature propagation on top of cross-view aggregation (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 53.72 -> 55.26 | Improved |
| add cross-sequence memory alignment (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 55.26 -> 56.65 | Improved |
| add cross-sequence memory alignment (ablation) | LiMA (ViT-B) · SemanticKITTI 1% | mIoU: 49.31 -> 50.44 | Improved |
| use average pooling instead of max pooling for cross-view aggregation (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 55.93 -> 56.65 | Improved |
| use average pooling instead of attention for cross-view aggregation (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 55.02 -> 56.65 | Improved |
| add average cross-view aggregation versus none (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 54.64 -> 56.65 | Improved |
| increase memory frames from 6 to 8 (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 56.65 -> 54.97 | Hurt |
| increase memory frames from 2 to 6 (ablation) | LiMA (ViT-B) · nuScenes LP | mIoU: 53.34 -> 56.65 | Improved |
| increase memory frames from 6 to 8 (ablation) | LiMA (ViT-B) · SemanticKITTI 1% | mIoU: 50.44 -> 51.78 | Mixed |
| use LiMA long-term distillation instead of ScaLR | LiMA · nuScenes LP (ViT-B) | mIoU: 51.90 -> 56.65 | Improved |
| use LiMA instead of ScaLR with CenterPoint | LiMA · nuScenes detection 5% | mAP: 44.3 -> 46.5 | Improved |
Closest papers by meaning
- Temporal Consistent 3D LiDAR Representation Learning for Semantic Perception in Autonomous Driving (CVPR 2023 · similarity 0.90)
- Enhanced Spatiotemporal Consistency for Image-to-LiDAR Data Pretraining (TPAMI 2025 · similarity 0.90)
- HilDA: Hierarchical Distillation with Diffusion for Advancing Self-Supervised LiDAR Pre-training (ECCV 2026 · similarity 0.90)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.89)
- LiMoE: Mixture of LiDAR Representation Learners from Automotive Scenes (CVPR 2025 · similarity 0.89)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (CVPR 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.
