mlsanityData to 14 Oct 2026

Probing Association Instability with Track-State Perturbations for Clip-Level Active Learning in Query-Propagation Multi-Object Tracking

arXiv 2608.17224 ↗First posted 18 Aug 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Training query-propagation end-to-end multi-object tracking (MOT) models requires dense bounding-box and identity annotations across video sequences, making dataset construction expensive. Clip-level active learning reduces this cost by selecting video clips for annotation, but prior acquisition criteria based on output-level temporal uncertainty may miss clips whose informativeness comes from association instability in propagated track states. We propose QPID (Query-Propagation Instability and Diversity), a clip acquisition method for query-propagation MOT that targets association instability in propagated track states. QPID estimates this instability by applying two-sided perturbations to internal track states and measuring prediction differences from a clean reference branch. The key idea is that, in stable clips, each propagated track should continue to follow the same target under small perturbations, whereas in ambiguous clips, small changes in the track state can alter which target the track follows, leading to changes in localization or confidence. QPID measures these perturbation-induced prediction differences with two metrics: Localization Drift and Entropy-Weighted Confidence Discrepancy. These metrics are aggregated into a clip-level association-instability score. To avoid redundant uncertainty-only selection, QPID selects a representative annotation batch from high-instability clips using Uncertainty-Weighted Visual Coverage with track-level visual prototypes. Experiments on DanceTrack and SportsMOT with MeMOTR and SambaMOTR show that QPID achieves strong performance compared with active learning baselines under the same annotation budget.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 42 of its 50 references that are in the map. First posted on arXiv 18 Aug 2026.

Builds on

42 of 50 references.

References · 42

  1. 2023Mamba: Linear-Time Sequence Modeling with Selective State Spacescited by 4,718
  2. 2019Objects as Pointscited by 1,175
  3. 2021YOLOX: Exceeding YOLO Series in 2021cited by 1,071
  4. 2016Simple Online and Realtime Trackingcited by 868
  5. 2021ByteTrack: Multi-Object Tracking by Associating Every Detection Boxcited by 823
  6. 2016Performance Measures and a Data Set for Multi-Target, Multi-Camera Trackingcited by 692
  7. 2019Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Boundscited by 610
  8. 2022DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETRcited by 573
  9. 2020HOTA: A Higher Order Metric for Evaluating Multi-Object Trackingcited by 552
  10. 2016MOT16: A Benchmark for Multi-Object Trackingcited by 532
  11. 2021TrackFormer: Multi-Object Tracking with Transformerscited by 518
  12. 2020FairMOT: On the Fairness of Detection and Re-Identification in Multiple Object Trackingcited by 490
30 more
  1. 2021MOTR: End-to-End Multiple-Object Tracking with Transformercited by 381
  2. 2020MOT20: A benchmark for multi object tracking in crowded scenescited by 357
  3. 2022Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Trackingcited by 346
  4. 2022BoT-SORT: Robust Associations Multi-Pedestrian Trackingcited by 320
  5. 2021DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motioncited by 217
  6. 2017Cost-Effective Active Learning for Deep Image Classificationcited by 166
  7. 2018Adversarial Active Learning for Deep Networks: a Margin Based Approachcited by 158
  8. 2022MOTRv2: Bootstrapping End-to-End Multi-Object Tracking by Pretrained Object Detectorscited by 148
  9. 2023SportsMOT: A Large Multi-Object Tracking Dataset in Multiple Sports Scenescited by 131
  10. 2021Batch Active Learning at Scalecited by 127
  11. 2023MeMOTR: Long-Term Memory-Augmented Transformer for Multi-Object Trackingcited by 88
  12. 2022Hard to Track Objects with Irregular Motions and Similar Appearances? Make It Easier by Buffering the Matching Spacecited by 66
  13. 2020Reinforced active learning for image segmentationcited by 56
  14. 2018Active Learning for Deep Object Detectioncited by 50
  15. 2018Localization-Aware Active Learning for Object Detectioncited by 48
  16. 2021Consistency-based Active Learning for Object Detectioncited by 32
  17. 2022Plug and Play Active Learning for Object Detectioncited by 28
  18. 2022Making Your First Choice: To Address Cold Start Problem in Vision Active Learningcited by 25
  19. 2023Active Prompt Learning in Vision Language Modelscited by 25
  20. 2019Temporal Coherence for Active Learning in Videoscited by 24
  21. 2024Samba: Synchronized Set-of-Sequences Modeling for Multiple Object Trackingcited by 21
  22. 2024Generalized Coverage for More Robust Low-Budget Active Learningcited by 16
  23. 2024Active Learning for Vision-Language Modelscited by 15
  24. 2024Uncertainty Herding: One Active Learning Method for All Label Budgetscited by 15
  25. 2023Best Practices in Active Learning for Semantic Segmentationcited by 10
  26. 2023Semi-supervised Active Learning for Video Action Detectioncited by 10
  27. 2024SPAMming Labels: Efficient Annotations for the Trackers of Tomorrowcited by 7
  28. 2023Correlation-aware active learning for surgery video segmentationcited by 4
  29. 2023Active Learning for Video Classification with Frame Level Queriescited by 2
  30. 2026Clip-level Uncertainty and Temporal-aware Active Learning for End-to-End Multi-Object Trackingcited by 1

Built on by

Not cited yet.