Research map / Detection and segmentation / Tracking
TAO: A Large-Scale Benchmark for Tracking Any Object
Achal Dave, Tarasha Khurana, Pavel Tokmakov, Cordelia Schmid, Deva Ramanan
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 oracle class labels to SORT tracks | SORT · TAO val | track mAP: 13.2 -> 30.2 | Improved |
| add a track oracle (perfect linking of detections) | SORT detections · TAO val | track mAP: 13.2 -> 31.5 | Improved |
| combine track oracle and class oracle | detections · TAO val | track mAP: 31.5 -> 83.6 | Improved |
| add re-id and frame alignment to Tracktor (ablation) | Tracktor++ · TAO val person | Person AP: 25.9 -> 36.7 | Improved |
| add re-id and camera motion compensation to Tracktor (ablation) | Tracktor · TAO val | MOTA: 61.6 -> 66.6 | Improved |
| add re-id and camera motion compensation to Tracktor (ablation) | Tracktor · MOT-17 train | MOTA: 61.5 -> 61.9 | No clear effect |
| apply Viterbi post-processing to a single-frame detector | Detector · ImageNet-Vid | detection mAP: 73.4 -> 79.2 | Improved |
| use SORT instead of Viterbi linking | Detector · TAO val | track mAP: 16.5 -> 18.5 (person AP) | Improved |
| initialize user trackers at largest box instead of first frame | DIMP* · TAO val | track mAP: 33.2 -> 38.5 | Improved |
| use user-initialized DIMP instead of SORT with class oracle | DIMP* · TAO val | MOTA: 11.3 -> 24.4 | Improved |
| lower mAP IoU threshold to 0.1 for user-init vs SORT | DIMP vs SORT · TAO val | track mAP: 36.9 -> 71.0 | Improved |
| lower SORT detection score threshold from 0.1 to 0.0005 (ablation) | SORT · TAO train | track mAP: 8.2 -> 11.3 | Improved |
Closest papers by meaning
- TrackingNet: A Large-Scale Dataset and Benchmark for Object Tracking in the Wild (ECCV 2018 · similarity 0.95)
- GOT-10k: A Large High-Diversity Benchmark for Generic Object Tracking in the Wild (TPAMI 2019 · similarity 0.92)
- COVTrack++: Learning Open-Vocabulary Multi-Object Tracking from Continuous Videos via a Synergistic Paradigm (TPAMI 2026 · similarity 0.92)
- OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking (NeurIPS 2024 · similarity 0.91)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.90)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.90)
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.
