Research map / Detection and segmentation / Tracking
MOTChallenge: A Benchmark for Single-Camera Multiple Target Tracking
Patrick Dendorfer, Aljoša Ošep, Anton Milan, Konrad Schindler B, Daniel Cremers, Ian D. Reid, Stefan Roth, Laura Leal-Taixé
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 |
|---|---|---|---|
| combine Tracktor-style regression with a learned message-passing graph solver | Tracktor++v2 (MPNTrack) · MOT15 (public detections, test) | MOTA: 46.60 -> 51.54 | Improved |
| combine Tracktor-style regression with a learned message-passing graph solver | Tracktor++v2 (MPNTrack) · MOT16 (public detections, test) | MOTA: 56.20 -> 58.56 | Improved |
| combine Tracktor-style regression with a learned message-passing graph solver | Tracktor++v2 (MPNTrack) · MOT17 (public detections, test) | MOTA: 56.35 -> 58.85 | Improved |
| fine-tune CNN appearance embeddings on target data instead of learned linear costs only | LP_SSVM (SiameseCNN) · MOT15 (public detections, test) | MOTA: 25.22 -> 29.04 | Improved |
| learn track management policies with reinforcement learning | hand-crafted affinity trackers (MDP) · MOT15 (public detections, test) | MOTA: 30.31 | Improved |
| use pre-trained CNN features with online-adapted appearance (MHT_DAM) instead of MDP | MDP · MOT15 (public detections, test) | MOTA: 30.31 -> 32.36 | Improved |
| add long-term interest point trajectories to affinity (NOMT) | hand-crafted trackers · MOT16 (public detections, test) | MOTA: 46.42 | Improved |
| replace multi-output least squares with bi-linear LSTM online adaptation | MHT_DAM · MOT17 (public detections, test) | MOTA: 50.71 -> 47.52 | Hurt |
| add a body joint detector to affinity (JBNOT) | tracking-by-detection · MOT17 (public detections, test) | MOTA: 52.63 | Improved |
| use motion-based and simple-distance association (SMOT, CEM, TBD, ELP, MotiCon) | tracking-by-detection · MOT16 (public detections, test) | MOTA: around 38% | No clear effect |
| PHD filter based joint state and association estimation | tracking-by-detection · MOT17 (public detections, test) | MOTA: 49.94 | No clear effect |
| multicut formulation without NMS pre-processing | tracking-by-detection · MOT16 (public detections, test) | MOTA: 48.78 and 49.25 | Improved |
Closest papers by meaning
- TAO: A Large-Scale Benchmark for Tracking Any Object (ECCV 2020 · similarity 0.92)
- GMOT-40: A Benchmark for Generic Multiple Object Tracking (CVPR 2021 · similarity 0.91)
- LaSOT: A High-quality Large-scale Single Object Tracking Benchmark (IJCV 2020 · similarity 0.91)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.90)
- DiffusionTrack: Diffusion Model for Multi-Object Tracking (AAAI 2024 · similarity 0.89)
- OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking (NeurIPS 2024 · 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.
