Research map / Point clouds / Point cloud classification
Segment Any Point Cloud Sequences by Distilling Vision Foundation Models
Youquan Liu, Lingdong Kong, Jun CEN, Runnan Chen, Wenwei Zhang, Liang Pan, Kai Chen, Ziwei 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 |
|---|---|---|---|
| use VFM (SEEM) semantic superpixels instead of SLIC (ablation) | Seal · nuScenes linear probing | mIoU: 40.89 -> 44.95 | Improved |
| use VFM superpixels (SEEM) in SLidR (ablation) | SLidR · nuScenes linear probing | mIoU: 38.80 -> 43.00 | Improved |
| add VFM superpixels to camera-to-LiDAR distillation (ablation) | Seal · nuScenes linear probing | mIoU: 38.80 -> 43.00 | Improved |
| add superpoint temporal consistency to C2L distillation (ablation) | Seal · nuScenes linear probing | mIoU: 38.80 -> 40.45 | Improved |
| add temporal consistency on top of VFM (ablation) | Seal · nuScenes linear probing | mIoU: 43.00 -> 44.01 | Improved |
| add point-to-segment regularization on top of VFM (ablation) | Seal · nuScenes linear probing | mIoU: 43.00 -> 43.35 | Improved |
| add point-to-segment regularization to VFM+STC (full model) (ablation) | Seal · nuScenes linear probing | mIoU: 44.01 -> 44.95 | Improved |
| add temporal consistency or P2S when fully fine-tuning on nuScenes (ablation) | Seal · nuScenes full fine-tune | mIoU: 75.21 -> 75.60 | No clear effect |
| add temporal consistency to VFM at 5% nuScenes fine-tune (ablation) | Seal · nuScenes 5% fine-tune | mIoU: 53.03 -> 55.36 | Improved |
| use SAM instead of SEEM superpixels (ablation) | Seal · nuScenes full fine-tune | mIoU: 75.60 -> 75.92 | Improved |
| use SAM instead of SEEM superpixels (ablation) | Seal · nuScenes linear probing | mIoU: 44.95 -> 43.94 | Hurt |
| Seal pretraining instead of SLidR under 10% sensor misalignment | SLidR · nuScenes 1% fine-tune | mIoU: 37.30 -> 44.80 | Improved |
Closest papers by meaning
- PointDistiller: Structured Knowledge Distillation Towards Efficient and Compact 3D Detection (CVPR 2023 · similarity 0.88)
- Learning Temporal Variations for 4D Point Cloud Segmentation (IJCV 2024 · similarity 0.87)
- Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Data (CVPR 2022 · similarity 0.87)
- CPCM: Contextual Point Cloud Modeling for Weakly-supervised Point Cloud Semantic Segmentation (ICCV 2023 · similarity 0.87)
- Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis (CVPR 2025 · similarity 0.86)
- VLM2Scene: Self-Supervised Image-Text-LiDAR Learning with Foundation Models for Autonomous Driving Scene Understanding (AAAI 2024 · similarity 0.86)
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.
