Research map / Point clouds / Point cloud classification
Boosting 3D Single Object Tracking with 2D Matching Distillation and 3D Pre-training
Qiangqiang Wu, Yan Xia, Jia Wan, Antoni Chan
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 Point-MAE 3D pre-training to SiamBase (ablation) | SiamBase · KITTI Van | Precision: 64.3 -> 69.3 | Improved |
| add 2D matching distillation to SiamBase with 3D pre-training (ablation) | SiamBase · KITTI Van | Precision: 69.3 -> 75.0 | Improved |
| add 2D matching distillation to MemBase with 3D pre-training (ablation) | MemBase · KITTI Van | Precision: 76.5 -> 79.3 | Improved |
| add IoU weighting to matching distillation (ablation) | SiamDisst · KITTI Van | Success/Precision: 56.4/66.3 -> 63.5/75.0 | Improved |
| use learnable projection in TAP instead of z-axis values (ablation) | SiamDisst · KITTI Van | Success/Precision: 61.1/72.3 -> 63.5/75.0 | Improved |
| add target-aware cross-attention in TAP (ablation) | SiamDisst · KITTI Van | Success/Precision: 61.2/73.1 -> 63.5/75.0 | Improved |
| use Point-MAE rather than ReCon, 2D MAE or DropMAE initialization (ablation) | SiamDisst · KITTI Van | Success/Precision: 60.5/69.9 (2D MAE) -> 63.5/75.0 | Improved |
| use heavier ViT-384-6 instead of ViT-96-3 matching module (ablation) | SiamDisst · KITTI Van | Success/Precision: 60.5/71.7 -> 63.5/75.0 | Improved |
| use heavier ViT-384-6 instead of ViT-96-3 matching module (ablation) | MemDisst · KITTI Van | Success/Precision: 66.6/79.3 vs 66.7/79.1 | No clear effect |
| apply matching distillation and 3D pre-training to STNet-style Siamese tracker | SiamDisst vs STNet · KITTI (Mean) | Success/Precision: 61.3/80.1 -> 66.2/83.9 | Improved |
| MemDisst versus MBPTrack on Waymo generalization | MemDisst · Waymo Open Dataset (Mean) | Success: 46.0 vs 45.9 | No clear effect |
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.
- CXTrack — CXTrack: Improving 3D Point Cloud Tracking With Contextual Information (CVPR 2023) · better on 1 of 1 shared result
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
- Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds (CVPR 2022 · similarity 0.91)
- TrackAny3D: Transferring Pretrained 3D Models for Category-unified 3D Point Cloud Tracking (ICCV 2025 · similarity 0.91)
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.89)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · similarity 0.88)
- Towards Category Unification of 3D Single Object Tracking on Point Clouds (ICLR 2024 · similarity 0.88)
- Modeling Continuous Motion for 3D Point Cloud Object Tracking (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.
