Research map / Point clouds / Point cloud classification
SPOT: Scalable 3D Pre-Training via Occupancy Prediction for Learning Transferable 3D Representations
Xiangchao Yan, Runjian Chen, Bo Zhang, Hancheng Ye, Renqiu Xia, Jiakang Yuan, Hongbin Zhou, Xinyu Cai, Botian Shi, Wenqi Shao, Ping Luo, Yu Qiao and 2 more
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 occupancy prediction pre-training (ablation) | SECOND · nuScenes (5% fine-tune) | mAP: 32.16 -> 36.55 | Improved |
| add loss balancing (ablation) | occupancy pre-training · nuScenes | mAP: 36.55 -> 37.90 | Improved |
| add beam re-sampling (ablation) | occupancy pre-training · nuScenes | mAP: 37.90 -> 38.63 | Improved |
| add dataset balancing (ablation) | occupancy pre-training · nuScenes | mAP: 38.63 -> 40.39 | Improved |
| add occupancy prediction (ablation) | SECOND · ONCE | mAP: 35.96 -> 36.00 | No clear effect |
| add beam re-sampling (ablation) | occupancy pre-training · ONCE | mAP: 37.30 -> 39.19 | Improved |
| use occupancy pre-training instead of detection pre-training (ablation) | CenterPoint · nuScenes det | mAP: 40.89 -> 47.47 | Improved |
| use occupancy pre-training instead of segmentation pre-training (ablation) | CenterPoint · nuScenes det | mAP: 36.23 -> 47.47 | Improved |
| use detection pre-training instead of no pre-training (ablation) | CenterPoint · nuScenes det | mAP: 42.37 -> 40.89 | Hurt |
| increase pre-training data from 5% to 100% (ablation) | SPOT SECOND · nuScenes (5% fine-tune) | mAP: 37.96 -> 42.57 | Improved |
| pre-train with weak labels only (no human labels) | SPOT SECOND · nuScenes (5% fine-tune) | mAP: 32.16 -> 42.24 | Improved |
| train from scratch for 150 epochs instead of 30 (ablation) | SECOND · nuScenes det | mAP: 32.16 -> 36.79 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BEV-MAE (100%) — BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PRED: Pre-training via Semantic Rendering on LiDAR Point Clouds (NeurIPS 2023 · similarity 0.92)
- Self-Supervised Pretraining for Large-Scale Point Clouds (NeurIPS 2022 · similarity 0.90)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.90)
- UnO: Unsupervised Occupancy Fields for Perception and Forecasting (CVPR 2024 · similarity 0.89)
- LiDAR-Net: A Real-scanned 3D Point Cloud Dataset for Indoor Scenes (CVPR 2024 · similarity 0.88)
- Is Your LiDAR Placement Optimized for 3D Scene Understanding? (NeurIPS 2024 · similarity 0.88)
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.
