Research map / Point clouds / Point cloud classification
TrackAny3D: Transferring Pretrained 3D Models for Category-unified 3D Point Cloud Tracking
Mengmeng Wang, Haonan Wang, Yulong Li, Xiangjie Kong, Jiaxin Du, Guojiang Shen, Feng Xia
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 |
|---|---|---|---|
| freeze RECON and add two-path adapters instead of full fine-tuning (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 62.0 -> 64.8 | Improved |
| add MoGE experts (ablation) | TrackAny3D adapter · KITTI (category-unified) | mean success: 64.8 -> 65.3 | Improved |
| add learnable temporal token (ablation) | TrackAny3D adapter+MoGE · KITTI (category-unified) | mean success: 65.3 -> 66.2 | Improved |
| add dynamic mask weight (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 66.2 -> 67.1 | Improved |
| add adapters to all layers instead of even layers (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 65.5 -> 67.1 | Improved |
| put adapters in all layers instead of last layer only (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 62.2 -> 67.1 | Improved |
| put MoGE in even layers instead of all layers (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 63.7 -> 67.1 | Improved |
| put MoGE in even layers instead of last layer only (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 61.7 -> 67.1 | Improved |
| use 3 frames for temporal propagation instead of 2 (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 64.0 -> 67.1 | Improved |
| use 3 frames instead of 5 (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 64.1 -> 67.1 | Improved |
| multiply mask by learnable beta instead of using raw mask (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 65.6 -> 67.1 | Improved |
| multiply mask by learnable beta instead of fully learnable mask (ablation) | TrackAny3D · KITTI (category-unified) | mean success: 66.1 -> 67.1 | 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.
- 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
- An Effective Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds (TPAMI 2023 · similarity 0.94)
- Modeling Continuous Motion for 3D Point Cloud Object Tracking (AAAI 2024 · similarity 0.92)
- Towards Category Unification of 3D Single Object Tracking on Point Clouds (ICLR 2024 · similarity 0.92)
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.91)
- UAWTrack: Universal 3D Single Object Tracking in Adverse Weather (AAAI 2025 · similarity 0.89)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · 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.
