Research map / Detection and segmentation / 3D detection and driving
TrajectoryFormer: 3D Object Tracking Transformer with Predictive Trajectory Hypotheses
Xuesong Chen, Shaoshuai Shi, Chao Zhang, Benjin Zhu, Qiang Wang, Ka Chun Cheung, Simon See, Hongsheng Li
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 one temporally predicted box to hypotheses (ablation) | TrajectoryFormer (CenterPoint) · Waymo val Vehicle | MOTA: 56.3 -> 59.5 | Improved |
| increase predicted boxes from 1 to 5 (ablation) | TrajectoryFormer (CenterPoint) · Waymo val Vehicle | MOTA: 59.5 -> 59.8 | Improved |
| increase predicted boxes from 5 to 10 (ablation) | TrajectoryFormer (CenterPoint) · Waymo val Vehicle | MOTA: 59.8 -> 59.7 | No clear effect |
| use refined detection box without predicted boxes vs CenterPoint (ablation) | TrajectoryFormer · Waymo val Vehicle | MOTA: 55.1 -> 56.3 | Improved |
| add short-term point appearance features to motion features (ablation) | TrajectoryFormer · Waymo val Vehicle | MOTA: 50.8 -> 59.8 | Improved |
| add long-term trajectory feature to point appearance feature (ablation) | TrajectoryFormer · Waymo val Vehicle | MOTA: 56.5 -> 59.8 | Improved |
| use 5 frames of points for appearance on pedestrians (ablation) | TrajectoryFormer · Waymo val Pedestrian | MOTA: 59.8 -> 61.0 | Improved |
| use multi-frame points for appearance on vehicles (ablation) | TrajectoryFormer · Waymo val Vehicle | MOTA: 59.6 -> 59.8 | No clear effect |
| increase history trajectory length from 5 to 10 frames (ablation) | TrajectoryFormer · Waymo val Vehicle | MOTA: 59.3 -> 59.8 | Improved |
| increase history trajectory length beyond 10 frames (ablation) | TrajectoryFormer · Waymo val Vehicle | MOTA: 59.8 -> 59.7 | No clear effect |
| use stronger detector MPPNet instead of CenterPoint | TrajectoryFormer · Waymo val Vehicle | MOTA: 59.7 -> 61.0 | Improved |
Later papers that compared with this method
1 later accepted paper reports TrajectoryFormer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Delving into Dynamic Scene Cue-Consistency for Robust 3D Multi-Object Tracking (AAAI 2026 · better on 3 of 3 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
- 3DMOTFormer: Graph Transformer for Online 3D Multi-Object Tracking (ICCV 2023 · similarity 0.93)
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · similarity 0.89)
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.89)
- GRAE-3DMOT: Geometry Relation-Aware Encoder for Online 3D Multi-Object Tracking (CVPR 2025 · similarity 0.88)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.88)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · 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.
