Research map / Point clouds / Point cloud classification
CLAP: Unsupervised 3D Representation Learning for Fusion 3D Perception via Curvature Sampling and Prototype Learning
Runjian Chen, Hang Zhang, Avinash Ravichandran, Hyoungseob Park, Wenqi Shao, Alex Wong, Ping Luo
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 joint pre-training with uniform range-guided sampling (ablation) | UniPAD · NuScenes (5% data) | mAP: 49.81 -> 49.55 | No clear effect |
| add Curvature Sampling to joint pre-training (ablation) | CLAP · NuScenes (5% data) | mAP: 49.55 -> 50.81 | Improved |
| add Prototype Learning (ablation) | CLAP · NuScenes (5% data) | mAP: 50.81 -> 51.17 | Improved |
| pre-train with CLAP instead of random init | BEVFusion · NuScenes (5% data) | mAP: 48.69 -> 51.17 | Improved |
| pre-train with ALSO instead of random init | BEVFusion · NuScenes (5% data) | mAP: 48.69 -> 45.34 | Hurt |
| pre-train with CLAP, fine-tune on 0.5% data | BEVFusion · NuScenes (0.5% data) | mAP: 16.49 -> 23.71 | Improved |
| pre-train with CLAP, fine-tune on 2.5% data | BEVFusion · NuScenes (2.5% data) | mAP: 39.12 -> 42.86 | Improved |
| pre-train with CLAP, fine-tune on 1% data | BEVFusion · NuScenes (1% data) | mAP: 26.22 -> 30.53 | Improved |
| pre-train with CLAP instead of random init | CenterPoint · Waymo (Level-1) | mAP: 61.60 -> 62.87 | Improved |
| pre-train with UniPAD instead of random init | CenterPoint · Waymo (Level-1) | mAP: 61.60 -> 61.57 | No clear effect |
| pre-train with CLAP instead of UniPAD | Cylinder3D · SemanticKITTI | mIoU: 31.88 -> 34.28 | Improved |
| pre-train on NuScenes with CLAP, transfer to ONCE | CenterPoint · ONCE (5% data) | mAP: 46.07 -> 46.88 | Improved |
Closest papers by meaning
- Clustering-based Supervised Training for 3D Perception (TPAMI 2026 · similarity 0.87)
- One at a Time: Progressive Multi-Step Volumetric Probability Learning for Reliable 3D Scene Perception (AAAI 2024 · similarity 0.87)
- FusionNeXt: Sequence-First 3D Multi-Modal Fusion in the Era of LLMs (NeurIPS 2026 · similarity 0.87)
- UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting (CVPR 2025 · similarity 0.86)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.86)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · 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.
