Research map / Detection and segmentation / Tracking
Association cues for online MOT
Improve detection-to-track association with appearance, ID prediction, and weak cues.
60 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Tracking, keyword cluster “mot · association · object tracking”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 4 of 6 tracked benchmarks.
15 papers in 2023–24 and 23 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 22.4% to 28.7%.
Open questions the data points to
- MOT17 test (IDF1): best went from 80.6 (up to 2024) to 83.1 in 2025 — Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · 31 citations)
- DanceTrack test (HOTA): best went from 65.7 (up to 2024) to 70.6 in 2025 — Foundation Model Driven Appearance Extraction for Robust Multiple Object Tracking (AAAI 2025 · 2 citations)
- MOT17 (MOTA): best went from 69.9 (up to 2024) to 82 in 2025 — More Than Meets the Eye: Enhancing Multi-Object Tracking Even with Prolonged Occlusions (ICML 2025)
- MOT17 (IDF1): best went from 70.1 (up to 2024) to 80.3 in 2025 — More Than Meets the Eye: Enhancing Multi-Object Tracking Even with Prolonged Occlusions (ICML 2025)
What looks settled
- MOT17 test (MOTA): best up to 2024 81.1, best in 2025–26 81.8 from 2 papers, less than 1% better.
- MOT20 test (MOTA): best up to 2024 78.1, best in 2025–26 78.3 from 2 papers, less than 1% better.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Online Multi-Object Tracking via Structural Constraint Event Aggregation (CVPR 2016 · 196 citations)
- 2019 · MOTS: Multi-Object Tracking and Segmentation (CVPR 2019 · 655 citations)
- 2020 · Learning a Neural Solver for Multiple Object Tracking (CVPR 2020 · 512 citations)
- 2021 · Learning a Proposal Classifier for Multiple Object Tracking (CVPR 2021 · 126 citations)
- 2022 · DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion (CVPR 2022 · 363 citations)
- 2023 · Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking (CVPR 2023 · 1,003 citations)
- 2024 · Hybrid-SORT: Weak Cues Matter for Online Multi-Object Tracking (AAAI 2024 · 173 citations)
- 2025 · Multiple Object Tracking as ID Prediction (CVPR 2025 · 61 citations)
- 2026 · When Trackers Date Fish: A Benchmark and Framework for Underwater Multiple Fish Tracking (AAAI 2026 · 5 citations)
Earliest papers here
- Online Multi-Object Tracking via Structural Constraint Event Aggregation (CVPR 2016 · 196 citations)
- MOTS: Multi-Object Tracking and Segmentation (CVPR 2019 · 655 citations)
- Eliminating Exposure Bias and Metric Mismatch in Multiple Object Tracking (CVPR 2019 · 67 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
MOT17 test · MOTA (higher is better)
9 papers, 2020–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 53.7 | How to Train Your Deep Multi-Object Tracker (CVPR 2020 · 229 citations) |
| 2021 | 71 | Improving Multiple Object Tracking With Single Object Tracking (CVPR 2021 · 123 citations) |
| 2023 | 81.1 | MotionTrack: Learning Robust Short-Term and Long-Term Motions for Multi-Object Tracking (CVPR 2023 · 168 citations) |
| 2025 | 81.8 | Foundation Model Driven Appearance Extraction for Robust Multiple Object Tracking (AAAI 2025 · 2 citations) |
MOT17 test · IDF1 (higher is better)
8 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 54.4 | A Unified Object Motion and Affinity Model for Online Multi-Object Tracking (CVPR 2020 · 120 citations) |
| 2021 | 71.9 | Improving Multiple Object Tracking With Single Object Tracking (CVPR 2021 · 123 citations) |
| 2023 | 80.1 | MotionTrack: Learning Robust Short-Term and Long-Term Motions for Multi-Object Tracking (CVPR 2023 · 168 citations) |
| 2024 | 80.6 | DeconfuseTrack: Dealing with Confusion for Multi-Object Tracking (CVPR 2024 · 24 citations) |
| 2025 | 83.1 | Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · 31 citations) |
MOT20 test · MOTA (higher is better)
7 papers, 2021–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 68.6 | Improving Multiple Object Tracking With Single Object Tracking (CVPR 2021 · 123 citations) |
| 2023 | 78 | MotionTrack: Learning Robust Short-Term and Long-Term Motions for Multi-Object Tracking (CVPR 2023 · 168 citations) |
| 2024 | 78.1 | DeconfuseTrack: Dealing with Confusion for Multi-Object Tracking (CVPR 2024 · 24 citations) |
| 2025 | 78.3 | Foundation Model Driven Appearance Extraction for Robust Multiple Object Tracking (AAAI 2025 · 2 citations) |
DanceTrack test · HOTA (higher is better)
6 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 54.6 | Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking (CVPR 2023 · 1,003 citations) |
| 2024 | 65.7 | Hybrid-SORT: Weak Cues Matter for Online Multi-Object Tracking (AAAI 2024 · 173 citations) |
| 2025 | 70.6 | Foundation Model Driven Appearance Extraction for Robust Multiple Object Tracking (AAAI 2025 · 2 citations) |
MOT17 · MOTA (higher is better)
6 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 44.2 | Eliminating Exposure Bias and Metric Mismatch in Multiple Object Tracking (CVPR 2019 · 67 citations) |
| 2020 | 63 | TubeTK: Adopting Tubes to Track Multi-Object in a One-Step Training Model (CVPR 2020 · 295 citations) |
| 2024 | 69.9 | Delving into the Trajectory Long-tail Distribution for Muti-object Tracking (CVPR 2024 · 28 citations) |
| 2025 | 82 | More Than Meets the Eye: Enhancing Multi-Object Tracking Even with Prolonged Occlusions (ICML 2025) |
MOT17 · IDF1 (higher is better)
6 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 61.7 | Learning a Neural Solver for Multiple Object Tracking (CVPR 2020 · 512 citations) |
| 2021 | 66.8 | Learning a Proposal Classifier for Multiple Object Tracking (CVPR 2021 · 126 citations) |
| 2024 | 70.1 | Delving into the Trajectory Long-tail Distribution for Muti-object Tracking (CVPR 2024 · 28 citations) |
| 2025 | 80.3 | More Than Meets the Eye: Enhancing Multi-Object Tracking Even with Prolonged Occlusions (ICML 2025) |
What has been tried
47 papers here report 503 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 5%, hurt 19%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add ROCM | DanceTrack-val · HOTA | 53.1 -> 53.7 | Helped | Hybrid-SORT: Weak Cues Matter for Online Multi-Object Tracking (AAAI 2024 · 173 citations) |
| Increase prior information proportion from 0% to 50% | MOT17 val-half · MOTA | 71.2 -> 74.5 | Helped | DiffusionTrack: Diffusion Model for Multi-Object Tracking (AAAI 2024 · 66 citations) |
| Add self-attention to the ID Decoder | DanceTrack · HOTA | 57.7 -> 59.7 | Helped | Multiple Object Tracking as ID Prediction (CVPR 2025 · 61 citations) |
| add multi-scale relation | BDD100K validation, setting #3 (quadruple downsampling) · HOTA | +2.0 HOTA | Helped | Towards Generalizable Multi-Object Tracking (CVPR 2024 · 36 citations) |
| increase training sample length from T=2 to T=3 | nuScenes validation · AMOTA | 0.346 -> 0.378 | Helped | ADA-Track: End-to-End Multi-Camera 3D Multi-Object Tracking with Alternating Detection and Association (CVPR 2024 · 32 citations) |
| add Track-Perspective-Based Association (TPA) | MOT17-val · HOTA | 67.3 -> 68.5 | Helped | Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · 31 citations) |
| Add SVA | MOT17 validation set · MOTA | 67.8 -> 68.3 | Helped | Delving into the Trajectory Long-tail Distribution for Muti-object Tracking (CVPR 2024 · 28 citations) |
| Add DDM alone | MOT17 val · HOTA | 69.0 -> 69.1 | Helped | DeconfuseTrack: Dealing with Confusion for Multi-Object Tracking (CVPR 2024 · 24 citations) |
| Use cascade attention KUM instead of visual-only feature unification | Refer-KITTI · HOTA | 39.09 -> 43.75 | Helped | iKUN: Speak to Trackers without Retraining (CVPR 2024 · 22 citations) |
| Add fine-grained Nets instead of coarse-grained SORT association | TAO novel · TETA | 28.1 -> 32.6 | Helped | NetTrack: Tracking Highly Dynamic Objects with a Net (CVPR 2024 · 21 citations) |
| Add Motion Margin Loss | Visdrone test set · MOTA | 39.6 -> 42.1 | Helped | MM-Tracker: Motion Mamba for UAV-platform Multiple Object Tracking (AAAI 2025 · 15 citations) |
| Add denoise instances | JRDB val · HOTA | 3.80 -> 24.32 | Helped | Omnidirectional Multi-Object Tracking (CVPR 2025 · 7 citations) |
What reviewers will expect
Datasets and baselines named in the 13 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- DanceTrack · 4 of 13 papers
- MOT17 · 3 of 13 papers
- MOT20 · 3 of 13 papers
- Refer-KITTI · 2 of 13 papers
- COCO · 2 of 13 papers
- nuScenes · 2 of 13 papers
Baselines
- ByteTrack · 6 of 13 papers
- OC-SORT · 6 of 13 papers
- FairMOT · 5 of 13 papers
- MOTRv2 · 4 of 13 papers
- MOTR · 4 of 13 papers
- TransRMOT · 2 of 13 papers
- TempRMOT · 2 of 13 papers
- BoT-SORT · 2 of 13 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 38 papers; from OpenAlex and the papers' affiliations).
- Huazhong University of Science and Technology · 4 papers
- Chinese Academy of Sciences · 3 papers
- Fudan University · 2 papers
- Tencent · 2 papers
- Carnegie Mellon University · 2 papers
- Zhejiang University · 2 papers
- City University of Hong Kong · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Memory and correlation-based MOT (52 papers)
- Unified transformer tracking (25 papers)
Elsewhere in the topic
- Transformer-based video object tracking (74 papers)
- Foundation-model and correlation-filter tracking (66 papers)
- Siamese network tracking (34 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).