Research map / Detection and segmentation / Tracking
Looking Beyond Two Frames: End-to-End Multi-Object Tracking Using Spatial and Temporal Transformers
Tianyu Zhu, Markus Hiller, Mahsa Ehsanpour, Rongkai Ma, Tom W. Drummond, Ian D. Reid, Hamid Rezatofighi
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 |
|---|---|---|---|
| increase track history length from 1 to 30 (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | MOTA: 55.4 -> 58.9 | Improved |
| increase track history length from 1 to 30 (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | IDF1: 48.4 -> 50.6 | Improved |
| increase track history length from 1 to 30 (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | FP: 4,700 -> 3,471 | Improved |
| increase track history length from 20 to 30 (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | ML: 19 -> 20 | Hurt |
| add false-negative augmentation (two frames) (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | MOTA: 12.2 -> 14.6 | Improved |
| add right-aligned insertion (two frames) (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | MOTA: 14.6 -> 28.4 | Improved |
| add false-positive augmentation (two frames) (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | MOTA: 28.4 -> 55.4 | Improved |
| add false-positive augmentation (30 frames) (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | MOTA: 39.2 -> 58.9 | Improved |
| train on 30 frames instead of two frames (full augmentation) (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | MOTA: 55.4 -> 58.9 | Improved |
| add false-negative augmentation to naive training (ablation) | MO3TR · Validation set (MOT20 + 2DMOT15) | IDF1: 22.1 -> 42.0 | Improved |
| predict initiation queries with multi-scale encoder (PIQ) (ablation) | MO3TR · MOT17 test (private detections) | MOTA: 63.9 -> 77.6 | Improved |
| predict initiation queries with multi-scale encoder (PIQ) | MO3TR · MOT17 test (private detections) | FPS: 11.9 -> 8.8 | Hurt |
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.
- FairMOTv2 — FairMOT: On the Fairness of Detection and Re-identification in Multiple Object Tracking (IJCV 2021) · 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
- TrackFormer: Multi-Object Tracking With Transformers (CVPR 2022 · similarity 0.91)
- MOTR: End-to-End Multiple-Object Tracking with TRansformer (ECCV 2022 · similarity 0.91)
- Standing Between Past and Future: Spatio-Temporal Modeling for Multi-Camera 3D Multi-Object Tracking (CVPR 2023 · similarity 0.91)
- MeMOTR: Long-Term Memory-Augmented Transformer for Multi-Object Tracking (ICCV 2023 · similarity 0.90)
- From Detection to Association: Learning Discriminative Object Embeddings for Multi-Object Tracking (CVPR 2026 · similarity 0.90)
- More Than Meets the Eye: Enhancing Multi-Object Tracking Even with Prolonged Occlusions (ICML 2025 · similarity 0.89)
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.
