Research map / Detection and segmentation / Tracking
How to Train Your Deep Multi-Object Tracker
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train Tracktor with the full DeepMOT loss | Tracktor · MOT17 test | MOTA: 53.5 -> 53.7 | Helped |
| Train Tracktor with the full DeepMOT loss | Tracktor · MOT16 test | MOTA: 54.4 -> 54.8 | Helped |
| Train Tracktor with the full DeepMOT loss | Tracktor · MOT15 test | MOTA: 44.1 -> 44.1 | No clear effect |
| Add a jointly trained ReID head | DeepMOT Tracktor · MOT17 validation | MOTA: 60.43 -> 60.66 | Helped |
| Use only dMOTP instead of the full loss (ablation) | Tracktor + ReIDhead · MOT17 validation | MOTA: 60.66 -> 60.51 | Hurt |
| Use only dMOTA instead of the full loss (ablation) | Tracktor + ReIDhead · MOT17 validation | MOTA: 60.66 -> 60.52 | Hurt |
| Remove the ID-switch penalty (ablation) | DeepMOT loss on Tracktor + ReIDhead · MOT17 validation | IDS: 124 -> 118 | Hurt |
| Replace GRU with LSTM units in the sequential DHN (ablation) | DHN · DHN test set, row-wise maximum | WA: 83.66 vs 92.88 | Hurt |
| Use parallel instead of sequential Bi-RNN processing (ablation) | DHN · DHN test set, row-wise maximum | WA: 89.56 vs 92.88 | Hurt |
| Replace the Bi-RNN with a 1D convolutional architecture (ablation) | DHN · DHN test set, row-wise maximum | WA: 56.43 vs 92.88 | Hurt |
| Train GOTURN with DeepMOT rather than Smooth L1 | GOTURN-based MOT · MOT17 validation | MOTA: 52.28 -> 54.09 | Helped |
| Train SiamRPN with DeepMOT rather than Smooth L1 | SiamRPN-based MOT · MOT17 validation | MOTA: 56.51 -> 57.16 | Helped |
Closest papers by meaning
- More Than Meets the Eye: Enhancing Multi-Object Tracking Even with Prolonged Occlusions (ICML 2025 · similarity 0.87)
- Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · similarity 0.86)
- Towards Generalizable Multi-Object Tracking (CVPR 2024 · similarity 0.85)
- Is Multiple Object Tracking a Matter of Specialization? (NeurIPS 2024 · similarity 0.85)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.85)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.85)
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.