pith. machine review for the scientific record. sign in

arxiv: 2605.09417 · v1 · submitted 2026-05-10 · 💻 cs.CV

Recognition: no theorem link

SAMOFT: Robust Multi-Object Tracking via Region and Flow

Yanchao Wang , Dawei Zhang , Chengzhuan Yang , Wei Liu , Minglu Li , Hua Wang , Zhonglong Zheng , Ming-Hsuan Yang
Authors on Pith no claims yet

Pith reviewed 2026-05-12 03:05 UTC · model grok-4.3

classification 💻 cs.CV
keywords multi-object trackingsegment anything modeloptical flowkalman filterre-identificationpixel-level cues
0
0 comments X

The pith

SAMOFT improves multi-object tracking by using pixel-level motion cues from the Segment Anything Model and optical flow to refine predictions under deformation and occlusion.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that instance-level features alone are insufficient for robust multi-object tracking when objects deform, move nonlinearly, or become occluded. It proposes combining the Segment Anything Model with dense optical flow to supply instantaneous foreground pixel motion and mask centroids that correct Kalman filter predictions and association decisions. Three specialized modules handle motion matching, low-confidence centroid alignment, and statistical correction of long-tailed trajectories, while a re-identification step stabilizes appearance cues. If these pixel-level additions work, trackers should maintain identities more reliably on challenging sequences without requiring extra training data for the motion components. A sympathetic reader cares because many real-world tracking applications, from surveillance to robotics, encounter exactly these failure modes.

Core claim

SAMOFT demonstrates that integrating SAM-derived masks with dense optical flow inside a Pixel Motion Matching module, a Centroid Distance Matching module, a Distribution-Based Correction module, and a Cluster-Aware ReID strategy produces more robust trajectory association than instance-level baselines alone, yielding consistent gains on DanceTrack and MOTChallenge benchmarks.

What carries the argument

The Pixel Motion Matching module, which fuses Segment Anything Model masks with dense optical flow to compute instantaneous foreground pixel motion and correct Kalman filter state estimates.

If this is right

  • Kalman filter motion predictions become more accurate without learning new motion models.
  • Low-confidence detections can still contribute to trajectories via mask centroid distances.
  • Historical flow statistics allow online correction of atypical motion without retraining.
  • Appearance features gain stability through clustering-aware re-identification.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same pixel-level correction idea could be grafted onto other association paradigms such as graph-based or transformer trackers.
  • If optical flow quality is the limiting factor, swapping in newer flow estimators might produce further gains on the same benchmarks.
  • The training-free distribution correction suggests that purely statistical motion models remain viable when paired with strong instantaneous cues.

Load-bearing premise

That the Segment Anything Model will produce usable foreground masks and that dense optical flow will supply accurate instantaneous motion even when objects deform, move nonlinearly, or are partially occluded.

What would settle it

Run SAMOFT on a sequence containing rapid nonlinear deformations and heavy occlusions where both SAM segmentation and optical flow visibly break down; if identity switches or track fragmentation exceed those of the unmodified baseline tracker, the pixel-cue benefit disappears.

Figures

Figures reproduced from arXiv: 2605.09417 by Chengzhuan Yang, Dawei Zhang, Hua Wang, Ming-Hsuan Yang, Minglu Li, Wei Liu, Yanchao Wang, Zhonglong Zheng.

Figure 1
Figure 1. Figure 1: (a) Pixel motion (i.e., optical flow magnitude) effectively captures [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall pipeline of SAMOFT. In the motion branch, PMM and CDM use SAM and an optical flow model to provide pixel-level matching cues, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of Pixel Motion Matching (PMM). Target pixel positions [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of mask centroids of a trajectory in the previous frame [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of HOTA, IDF1, and MOTA scores on the DanceTrack [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Tracking comparison between the baseline tracker and SAMOFT. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: SAM segmentation under partial occlusion. Despite significant overlap [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
read the original abstract

Multi-object tracking (MOT) is a fundamental task in computer vision that requires continuously tracking multiple targets while maintaining consistent identities across frames. However, most existing approaches primarily rely on instance-level object features for trajectory association, which often leads to degraded performance under challenging conditions such as object deformation, nonlinear motion, and occlusion. In this work, we propose SAMOFT, a robust tracker that leverages pixel-level cues to improve robustness under complex motion scenarios. Specifically, we introduce a Pixel Motion Matching (PMM) module that integrates the Segment Anything Model (SAM) with dense optical flow to refine Kalman filter-based motion prediction using instantaneous foreground pixel motion. To further enhance robustness under unreliable detections, we design a Centroid Distance Matching (CDM) module that performs flexible mask-based centroid matching for low-confidence or partially occluded observations. Moreover, a Distribution-Based Correction (DBC) module models long-tailed motion patterns in a training-free manner using historical optical flow statistics and dynamically corrects trajectory states online. We also incorporate a Cluster-Aware ReID (CA-ReID) strategy to improve the stability and discriminative power of trajectory appearance features. Extensive experiments on the DanceTrack and MOTChallenge benchmarks demonstrate that SAMOFT consistently improves baseline trackers and achieves competitive performance compared with recent state-of-the-art methods, validating the effectiveness of leveraging pixel-level cues for robust multi-object tracking.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper proposes SAMOFT, a multi-object tracker that augments Kalman-filter-based methods with pixel-level cues derived from the Segment Anything Model (SAM) and dense optical flow. It introduces a Pixel Motion Matching (PMM) module to refine motion predictions using instantaneous foreground pixel motion, a Centroid Distance Matching (CDM) module for mask-based centroid association under low-confidence or occluded detections, a Distribution-Based Correction (DBC) module that uses historical optical flow statistics to model and correct long-tailed motion patterns in a training-free manner, and a Cluster-Aware ReID (CA-ReID) strategy to enhance appearance feature stability. The central claim is that these components yield consistent improvements over baselines and competitive results against recent state-of-the-art methods on the DanceTrack and MOTChallenge benchmarks, validating the utility of pixel-level cues where instance-level features fail.

Significance. If the reported benchmark gains are supported by controlled ablations and the modules demonstrably address the targeted failure modes, the work would be significant for the MOT community. It provides a practical demonstration of integrating a foundation model (SAM) with classical motion models and flow without requiring retraining, and the training-free DBC component is a clear strength that could be adopted more broadly. The approach also offers a concrete path for hybrid region-and-flow trackers in deformation and occlusion scenarios.

major comments (2)
  1. [§3.1 and §3.2] §3.1 (PMM) and §3.2 (CDM): The central robustness claim rests on SAM producing reliable foreground masks and centroids under object deformation, nonlinear motion, and occlusion—the exact conditions the paper targets. No quantitative evaluation of SAM mask quality (e.g., IoU against ground-truth or failure rate on DanceTrack/MOTChallenge sequences) is provided, leaving open the possibility that observed gains derive primarily from the underlying SAM and flow models rather than the proposed matching logic.
  2. [§4] §4 Experiments: The abstract and results claim “consistent improvements” and “competitive performance,” yet the manuscript supplies no error bars, statistical significance tests, or per-sequence breakdowns. Without these, it is impossible to determine whether the reported deltas exceed typical tracker variance or are driven by a few easy sequences.
minor comments (3)
  1. [Figure 1] Figure 1 (overview): The diagram is too compressed; the flow arrows between PMM, CDM, and DBC are difficult to follow. Enlarging the figure and adding explicit labels for each data path would improve readability.
  2. [§3.3] Notation in §3.3 (DBC): The long-tailed distribution is described only qualitatively; a short equation or pseudocode showing how the historical flow histogram is maintained and used for online correction would remove ambiguity.
  3. [Related Work] Related work: Several recent MOT papers that also combine optical flow or mask cues (e.g., those building on RAFT or Mask2Former) are not cited, weakening the positioning of the novelty.

Simulated Author's Rebuttal

2 responses · 1 unresolved

Thank you for your detailed and constructive review. We appreciate the feedback on strengthening the empirical validation of our modules. We address each major comment below.

read point-by-point responses

  1. Referee: [§3.1 and §3.2] §3.1 (PMM) and §3.2 (CDM): The central robustness claim rests on SAM producing reliable foreground masks and centroids under object deformation, nonlinear motion, and occlusion—the exact conditions the paper targets. No quantitative evaluation of SAM mask quality (e.g., IoU against ground-truth or failure rate on DanceTrack/MOTChallenge sequences) is provided, leaving open the possibility that observed gains derive primarily from the underlying SAM and flow models rather than the proposed matching logic.

    Authors: We acknowledge the value of direct quantitative validation of SAM mask quality. However, DanceTrack and MOTChallenge provide only bounding-box annotations and lack pixel-level ground-truth masks, precluding IoU computation without new annotations. Our ablation studies (Table 3) isolate the contribution of PMM and CDM by comparing against baselines that use identical SAM and flow inputs, showing consistent metric gains attributable to the matching logic rather than the foundation models alone. In revision we will add a qualitative analysis of mask reliability under deformation/occlusion together with selected failure cases. revision: partial

  2. Referee: [§4] §4 Experiments: The abstract and results claim “consistent improvements” and “competitive performance,” yet the manuscript supplies no error bars, statistical significance tests, or per-sequence breakdowns. Without these, it is impossible to determine whether the reported deltas exceed typical tracker variance or are driven by a few easy sequences.

    Authors: We agree that error bars, significance testing, and per-sequence breakdowns would increase rigor. In the revised manuscript we will report standard deviations (where stochasticity exists, e.g., ReID clustering), per-sequence MOTA/IDF1 on DanceTrack and MOT17, and paired t-tests confirming that the observed deltas are statistically significant and consistent across sequences rather than driven by outliers. revision: yes

standing simulated objections not resolved
  • Quantitative IoU evaluation of SAM masks against ground truth cannot be performed because the standard MOT benchmarks supply only bounding-box annotations.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes an applied MOT system that composes external pretrained components (SAM, dense optical flow, Kalman filter) with four processing modules whose outputs are defined by independent image-processing operations rather than by the final tracking metric. No equations, parameter fits, or self-citations are shown to reduce the claimed benchmark gains to the inputs by construction; the validation rests on controlled experiments and ablations on public datasets whose ground truth is external to the method. The derivation is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Only the abstract is available, so the ledger is limited to assumptions stated or implied in the proposal. The work relies on the reliability of SAM segmentation and optical flow under the targeted conditions; no free parameters or new physical entities are explicitly introduced in the abstract.

axioms (2)
  • domain assumption SAM produces accurate pixel-level foreground masks suitable for motion matching
    Invoked in the Pixel Motion Matching module description.
  • domain assumption Dense optical flow provides instantaneous foreground pixel motion that improves Kalman predictions
    Central to both PMM and DBC modules.

pith-pipeline@v0.9.0 · 5562 in / 1425 out tokens · 78947 ms · 2026-05-12T03:05:14.498748+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

70 extracted references · 70 canonical work pages · 1 internal anchor

  1. [1]

    MOT16: A Benchmark for Multi-Object Tracking

    A. Milan, L. Leal-Taix ´e, I. Reid, S. Roth, and K. Schindler, “Mot16: A benchmark for multi-object tracking,”arXiv preprint arXiv:1603.00831,

    work page Pith review arXiv

  2. [2]

    Dancetrack: Multi-object tracking in uniform appearance and diverse motion,

    P. Sun, J. Cao, Y . Jiang, Z. Yuan, S. Bai, K. Kitani, and P. Luo, “Dancetrack: Multi-object tracking in uniform appearance and diverse motion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 20 993–21 002. 1, 2, 7

    work page 2022

  3. [3]

    Simple online and realtime tracking,

    A. Bewley, Z. Ge, L. Ott, F. Ramos, and B. Upcroft, “Simple online and realtime tracking,” inIEEE International Conference on Image Processing, 2016, pp. 3464–3468. 1, 2

    work page 2016

  4. [4]

    Observation- centric sort: Rethinking sort for robust multi-object tracking,

    J. Cao, J. Pang, X. Weng, R. Khirodkar, and K. Kitani, “Observation- centric sort: Rethinking sort for robust multi-object tracking,” inPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 9686–9696. 1, 2, 3, 6, 7, 8

    work page 2023

  5. [5]

    Hybrid-sort: Weak cues matter for online multi-object tracking,

    M. Yang, G. Han, B. Yan, W. Zhang, J. Qi, H. Lu, and D. Wang, “Hybrid-sort: Weak cues matter for online multi-object tracking,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 7, 2024, pp. 6504–6512. 1, 3, 6, 7, 8

    work page 2024

  6. [6]

    YOLOX: Exceeding YOLO Series in 2021

    Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, “Yolox: Exceeding yolo series in 2021,”arXiv preprint arXiv:2107.08430, 2021. 1, 2, 6

    work page internal anchor Pith review arXiv 2021

  7. [7]

    Bytetrack: Multi-object tracking by associating every detection box,

    Y . Zhang, P. Sun, Y . Jiang, D. Yu, F. Weng, Z. Yuan, P. Luo, W. Liu, and X. Wang, “Bytetrack: Multi-object tracking by associating every detection box,” inEuropean Conference on Computer Vision, 2022, pp. 1–21. 1, 2, 3, 5, 6, 7, 8, 9

    work page 2022

  8. [8]

    Deep oc-sort: Multi- pedestrian tracking by adaptive re-identification,

    G. Maggiolino, A. Ahmad, J. Cao, and K. Kitani, “Deep oc-sort: Multi- pedestrian tracking by adaptive re-identification,” inIEEE International Conference on Image Processing, 2023, pp. 3025–3029. 1, 3, 6, 7

    work page 2023

  9. [9]

    Quo vadis: Is trajectory forecasting the key towards long-term multi-object track- ing?

    P. Dendorfer, V . Yugay, A. Osep, and L. Leal-Taix ´e, “Quo vadis: Is trajectory forecasting the key towards long-term multi-object track- ing?”Advances in Neural Information Processing Systems, vol. 35, pp. 15 657–15 671, 2022. 1, 3

    work page 2022

  10. [10]

    Thermal pedestrian multiple object tracking challenge (tp-mot),

    W. El Ahmar, A. Sappa, and R. Hammoud, “Thermal pedestrian multiple object tracking challenge (tp-mot),” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 4602–4609. 1

    work page 2025

  11. [11]

    Visible–thermal multiple object tracking: Large-scale video dataset and progressive fusion approach,

    Y . Zhu, Q. Wang, C. Li, J. Tang, C. Gu, and Z. Huang, “Visible–thermal multiple object tracking: Large-scale video dataset and progressive fusion approach,”Pattern Recognition, vol. 161, p. 111330, 2025. 1

    work page 2025

  12. [12]

    Multi-target multi- camera tracking by tracklet-to-target assignment,

    Y . He, X. Wei, X. Hong, W. Shi, and Y . Gong, “Multi-target multi- camera tracking by tracklet-to-target assignment,”IEEE Transactions on Image Processing, vol. 29, pp. 5191–5205, 2020. 1

    work page 2020

  13. [13]

    Omnidirectional multi-object tracking,

    K. Luo, H. Shi, S. Wu, F. Teng, M. Duan, C. Huang, Y . Wang, K. Wang, and K. Yang, “Omnidirectional multi-object tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 21 959–21 969. 1

    work page 2025

  14. [14]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 4015–4026. 2, 3, 6

    work page 2023

  15. [15]

    arXiv preprint arXiv:2003.09003 (2020)

    P. Dendorfer, H. Rezatofighi, A. Milan, J. Shi, D. Cremers, I. Reid, S. Roth, K. Schindler, and L. Leal-Taix ´e, “Mot20: A bench- mark for multi object tracking in crowded scenes,”arXiv preprint arXiv:2003.09003, 2020. 2

    work page arXiv 2003

  16. [16]

    Motr: End-to-end multiple-object tracking with transformer,

    F. Zeng, B. Dong, Y . Zhang, T. Wang, X. Zhang, and Y . Wei, “Motr: End-to-end multiple-object tracking with transformer,” inEuropean Conference on Computer Vision, 2022, pp. 659–675. 2, 7, 8

    work page 2022

  17. [17]

    Motrv2: Bootstrapping end-to-end multi-object tracking by pretrained object detectors,

    Y . Zhang, T. Wang, and X. Zhang, “Motrv2: Bootstrapping end-to-end multi-object tracking by pretrained object detectors,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 056–22 065. 2, 7, 8

    work page 2023

  18. [18]

    Inference-domain network evolution: A new perspective for one-shot multi-object tracking,

    R. Li, B. Zhang, J. Liu, W. Liu, and Z. Teng, “Inference-domain network evolution: A new perspective for one-shot multi-object tracking,”IEEE Transactions on Image Processing, vol. 32, pp. 2147–2159, 2023. 2

    work page 2023

  19. [19]

    Pro2diff: Proposal propagation for multi-object tracking via the diffusion model,

    H. Liu, C. Zhang, B. Fan, and J. Xu, “Pro2diff: Proposal propagation for multi-object tracking via the diffusion model,”IEEE Transactions on Image Processing, vol. 33, pp. 6508–6520, 2024. 2

    work page 2024

  20. [20]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,”Advances in Neural Information Processing Systems, vol. 28, 2015. 2

    work page 2015

  21. [21]

    Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,

    C.-Y . Wang, A. Bochkovskiy, and H.-Y . M. Liao, “Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 7464–7475. 2

    work page 2023

  22. [22]

    The hungarian method for the assignment problem,

    H. W. Kuhn, “The hungarian method for the assignment problem,”Naval Research Logistics Quarterly, vol. 2, no. 1-2, pp. 83–97, 1955. 2

    work page 1955

  23. [23]

    Contributions to the theory of optimal control,

    R. E. Kalmanet al., “Contributions to the theory of optimal control,” Bol. Soc. Mat. Mexicana, vol. 5, no. 2, pp. 102–119, 1960. 3

    work page 1960

  24. [24]

    Strong- sort: Make deepsort great again,

    Y . Du, Z. Zhao, Y . Song, Y . Zhao, F. Su, T. Gong, and H. Meng, “Strong- sort: Make deepsort great again,”IEEE Transactions on Multimedia, vol. 25, pp. 8725–8737, 2023. 3, 7, 8

    work page 2023

  25. [25]

    Conftrack: Kalman filter-based multi-person tracking by utilizing confidence score of detection box,

    H. Jung, S. Kang, T. Kim, and H. Kim, “Conftrack: Kalman filter-based multi-person tracking by utilizing confidence score of detection box,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 6583–6592. 3

    work page 2024

  26. [26]

    Bpmtrack: Multi-object tracking with detection box application pattern mining,

    Y . Gao, H. Xu, J. Li, and X. Gao, “Bpmtrack: Multi-object tracking with detection box application pattern mining,”IEEE Transactions on Image Processing, vol. 33, pp. 1508–1521, 2024. 3

    work page 2024

  27. [27]

    Sampling-resilient multi-object tracking,

    Z. Li, D. Zhang, S. Wu, M. Song, and G. Chen, “Sampling-resilient multi-object tracking,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 4, 2024, pp. 3297–3305. 3

    work page 2024

  28. [28]

    ikun: Speak to trackers without retraining,

    Y . Du, C. Lei, Z. Zhao, and F. Su, “ikun: Speak to trackers without retraining,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 135–19 144. 3

    work page 2024

  29. [29]

    Diffmot: A real-time diffusion-based multiple object tracker with non-linear prediction,

    W. Lv, Y . Huang, N. Zhang, R.-S. Lin, M. Han, and D. Zeng, “Diffmot: A real-time diffusion-based multiple object tracker with non-linear prediction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 321–19 330. 3, 7

    work page 2024

  30. [30]

    Simple online and realtime track- ing with a deep association metric,

    N. Wojke, A. Bewley, and D. Paulus, “Simple online and realtime track- ing with a deep association metric,” inIEEE International Conference on Image Processing, 2017, pp. 3645–3649. 3

    work page 2017

  31. [31]

    Towards real-time multi-object tracking,

    Z. Wang, L. Zheng, Y . Liu, Y . Li, and S. Wang, “Towards real-time multi-object tracking,” inEuropean Conference on Computer Vision, 2020, pp. 107–122. 3

    work page 2020

  32. [32]

    Bot-sort: R obust associations multi-pedestrian tracking

    N. Aharon, R. Orfaig, and B.-Z. Bobrovsky, “Bot-sort: Robust associa- tions multi-pedestrian tracking,”arXiv preprint arXiv:2206.14651, 2022. 3, 6 12

    work page arXiv 2022

  33. [33]

    Simple cues lead to a strong multi-object tracker,

    J. Seidenschwarz, G. Bras ´o, V . C. Serrano, I. Elezi, and L. Leal-Taix ´e, “Simple cues lead to a strong multi-object tracker,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 13 813–13 823. 3, 7, 8

    work page 2023

  34. [34]

    Stat: Multi-object tracking based on spatio-temporal topological constraints,

    J. Zhang, M. Wang, H. Jiang, X. Zhang, C. Yan, and D. Zeng, “Stat: Multi-object tracking based on spatio-temporal topological constraints,” IEEE Transactions on Multimedia, vol. 26, pp. 4445–4457, 2023. 3, 7, 8

    work page 2023

  35. [35]

    Topic: A parallel association paradigm for multi-object tracking under complex motions and diverse scenes,

    X. Cao, Y . Zheng, Y . Yao, H. Qin, X. Cao, and S. Guo, “Topic: A parallel association paradigm for multi-object tracking under complex motions and diverse scenes,”IEEE Transactions on Image Processing, vol. 34, pp. 743–758, 2025. 3, 7, 8

    work page 2025

  36. [36]

    Focusing on tracks for online multi-object tracking,

    K. Shim, K. Ko, Y . Yang, and C. Kim, “Focusing on tracks for online multi-object tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 11 687–11 696. 3, 7

    work page 2025

  37. [37]

    Deconfusetrack: Dealing with confusion for multi-object tracking,

    C. Huang, S. Han, M. He, W. Zheng, and Y . Wei, “Deconfusetrack: Dealing with confusion for multi-object tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 290–19 299. 3, 7, 8

    work page 2024

  38. [38]

    Sparsetrack: Multi- object tracking by performing scene decomposition based on pseudo- depth,

    Z. Liu, X. Wang, C. Wang, W. Liu, and X. Bai, “Sparsetrack: Multi- object tracking by performing scene decomposition based on pseudo- depth,”IEEE Transactions on Circuits and Systems for Video Technol- ogy, vol. 35, no. 5, pp. 4870–4882, 2025. 3

    work page 2025

  39. [39]

    Deep kalman filter with optical flow for multiple object tracking,

    Y . Chen, D. Zhao, and H. Li, “Deep kalman filter with optical flow for multiple object tracking,” in2019 IEEE International Conference on Systems, Man and Cybernetics (SMC), 2019, pp. 3036–3041. 3

    work page 2019

  40. [40]

    Folt: Fast multiple object tracking from uav-captured videos based on optical flow,

    M. Yao, J. Wang, J. Peng, M. Chi, and C. Liu, “Folt: Fast multiple object tracking from uav-captured videos based on optical flow,” inProceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 3375–3383. 3

    work page 2023

  41. [41]

    Temporal coherent object flow for multi-object tracking,

    Z. Song, R. Luo, L. Ma, Y . Tang, Y .-P. P. Chen, J. Yu, and W. Yang, “Temporal coherent object flow for multi-object tracking,” inProceed- ings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 7, 2025, pp. 6978–6986. 3, 7, 8

    work page 2025

  42. [42]

    Mode-track: Robust multi-object tracking with motion decoupling in uav videos,

    Z. Song, Y . Li, S. Zhou, W. Tang, and L. Wang, “Mode-track: Robust multi-object tracking with motion decoupling in uav videos,”IEEE Transactions on Multimedia, pp. 1–11, 2026. 3

    work page 2026

  43. [43]

    Samu- rai: Motion-aware memory for training-free visual object tracking with sam 2,

    C.-Y . Yang, H.-W. Huang, W. Chai, Z. Jiang, and J.-N. Hwang, “Samu- rai: Motion-aware memory for training-free visual object tracking with sam 2,”IEEE Transactions on Image Processing, vol. 35, pp. 970–982,

    work page

  44. [44]

    Matching anything by segmenting anything,

    S. Li, L. Ke, M. Danelljan, L. Piccinelli, M. Segu, L. Van Gool, and F. Yu, “Matching anything by segmenting anything,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 18 963–18 973. 3

    work page 2024

  45. [45]

    Sam2mot: A novel paradigm of multi-object tracking by segmentation,

    J. Jiang, Z. Wang, M. Zhao, Y . Li, and D. Jiang, “Sam2mot: A novel paradigm of multi-object tracking by segmentation,”arXiv preprint arXiv:2504.04519, 2025. 3

    work page arXiv 2025

  46. [46]

    Delving into the trajectory long-tail distribution for muti-object tracking,

    S. Chen, E. Yu, J. Li, and W. Tao, “Delving into the trajectory long-tail distribution for muti-object tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 341–19 351. 3

    work page 2024

  47. [47]

    Nettrack: Tracking highly dynamic objects with a net,

    G. Zheng, S. Lin, H. Zuo, C. Fu, and J. Pan, “Nettrack: Tracking highly dynamic objects with a net,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 19 145–19 155. 3

    work page 2024

  48. [48]

    Cotracker: It is better to track together,

    N. Karaev, I. Rocco, B. Graham, N. Neverova, A. Vedaldi, and C. Rup- precht, “Cotracker: It is better to track together,” inEuropean Conference on Computer Vision, 2024, pp. 18–35. 3

    work page 2024

  49. [49]

    Sea-raft: Simple, efficient, accurate raft for optical flow,

    Y . Wang, L. Lipson, and J. Deng, “Sea-raft: Simple, efficient, accurate raft for optical flow,” inEuropean Conference on Computer Vision, 2024, pp. 36–54. 3, 6

    work page 2024

  50. [50]

    Hota: A higher order metric for evaluating multi-object tracking,

    J. Luiten, A. Osep, P. Dendorfer, P. Torr, A. Geiger, L. Leal-Taix ´e, and B. Leibe, “Hota: A higher order metric for evaluating multi-object tracking,”International Journal of Computer Vision, vol. 129, no. 2, pp. 548–578, 2021. 6

    work page 2021

  51. [51]

    Performance measures and a data set for multi-target, multi-camera tracking,

    E. Ristani, F. Solera, R. Zou, R. Cucchiara, and C. Tomasi, “Performance measures and a data set for multi-target, multi-camera tracking,” in European Conference on Computer Vision, 2016, pp. 17–35. 6

    work page 2016

  52. [52]

    Evaluating multiple object tracking performance: The clear mot metrics,

    K. Bernardin and R. Stiefelhagen, “Evaluating multiple object tracking performance: The clear mot metrics,”EURASIP Journal on Image and Video Processing, vol. 2008, no. 1, p. 246309, 2008. 6

    work page 2008

  53. [53]

    Global tracking transformers,

    X. Zhou, T. Yin, V . Koltun, and P. Kr ¨ahenb¨uhl, “Global tracking transformers,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 8771–8780. 7, 8

    work page 2022

  54. [54]

    Unifying short and long- term tracking with graph hierarchies,

    O. Cetintas, G. Bras ´o, and L. Leal-Taix ´e, “Unifying short and long- term tracking with graph hierarchies,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 877–22 887. 7, 8

    work page 2023

  55. [55]

    Diffusiontrack: Diffusion model for multi-object tracking,

    R. Luo, Z. Song, L. Ma, J. Wei, W. Yang, and M. Yang, “Diffusiontrack: Diffusion model for multi-object tracking,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 5, 2024, pp. 3991–

    work page 2024

  56. [56]

    Multiple object tracking as id prediction,

    R. Gao, J. Qi, and L. Wang, “Multiple object tracking as id prediction,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 27 883–27 893. 7, 8

    work page 2025

  57. [57]

    Focus on details: Online multi-object tracking with diverse fine-grained repre- sentation,

    H. Ren, S. Han, H. Ding, Z. Zhang, H. Wang, and F. Wang, “Focus on details: Online multi-object tracking with diverse fine-grained repre- sentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 11 289–11 298. 7

    work page 2023

  58. [58]

    Ahor: Online multi- object tracking with authenticity hierarchizing and occlusion recovery,

    H. Jin, X. Nie, Y . Yan, X. Chen, Z. Zhu, and D. Qi, “Ahor: Online multi- object tracking with authenticity hierarchizing and occlusion recovery,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 9, pp. 8253–8265, 2024. 7

    work page 2024

  59. [59]

    A confidence-aware matching strategy for generalized multi-object tracking,

    K. Shim, J. Hwang, K. Ko, and C. Kim, “A confidence-aware matching strategy for generalized multi-object tracking,” inIEEE International Conference on Image Processing, 2024, pp. 4042–4048. 7

    work page 2024

  60. [60]

    Towards generalizable multi-object tracking,

    Z. Qin, L. Wang, S. Zhou, P. Fu, G. Hua, and W. Tang, “Towards generalizable multi-object tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 18 995–19 004. 7, 8

    work page 2024

  61. [61]

    Ucmc- track: Multi-object tracking with uniform camera motion compensation,

    K. Yi, K. Luo, X. Luo, J. Huang, H. Wu, R. Hu, and W. Hao, “Ucmc- track: Multi-object tracking with uniform camera motion compensation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 7, 2024, pp. 6702–6710. 7

    work page 2024

  62. [62]

    Multi-scene generalized trajectory global graph solver with composite nodes for multiple object tracking,

    Y . Gao, H. Xu, J. Li, N. Wang, and X. Gao, “Multi-scene generalized trajectory global graph solver with composite nodes for multiple object tracking,” inProceedings of the AAAI Conference on Artificial Intelli- gence, vol. 38, no. 3, 2024, pp. 1842–1850. 7

    work page 2024

  63. [63]

    Dftrack: Deconfused data association framework for multi-object tracking,

    C. Huang, S. Han, M. He, W. Zheng, and Y . Wei, “Dftrack: Deconfused data association framework for multi-object tracking,”IEEE Transac- tions on Circuits and Systems for Video Technology, vol. 35, no. 10, pp. 10 367–10 381, 2025. 7

    work page 2025

  64. [64]

    Smiletrack: Similarity learning for occlusion-aware multiple object tracking,

    Y .-H. Wang, J.-W. Hsieh, P.-Y . Chen, M.-C. Chang, H.-H. So, and X. Li, “Smiletrack: Similarity learning for occlusion-aware multiple object tracking,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 6, 2024, pp. 5740–5748. 8

    work page 2024

  65. [65]

    Tgformer: Transformer with track query group for multi-object tracking,

    R. Zeng, Y . Huang, and S. Pei, “Tgformer: Transformer with track query group for multi-object tracking,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 9, 2025, pp. 9824–9832. 8

    work page 2025

  66. [66]

    Foundation model driven ap- pearance extraction for robust multiple object tracking,

    T. Fu, H. Yu, K. Niu, B. Li, and X. Xue, “Foundation model driven ap- pearance extraction for robust multiple object tracking,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 3, 2025, pp. 3031–3039. 8

    work page 2025

  67. [67]

    La-motr: End-to- end multi-object tracking by learnable association,

    P. Wang, Y . Wang, H. Cao, W. Chen, and D. Li, “La-motr: End-to- end multi-object tracking by learnable association,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 12 438–12 448. 8

    work page 2025

  68. [68]

    Motiontrack: Learning robust short-term and long-term motions for multi-object tracking,

    Z. Qin, S. Zhou, L. Wang, J. Duan, G. Hua, and W. Tang, “Motiontrack: Learning robust short-term and long-term motions for multi-object tracking,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 17 939–17 948. 7, 8

    work page 2023

  69. [69]

    Spring: A high-resolution high-detail dataset and benchmark for scene flow, optical flow and stereo,

    L. Mehl, J. Schmalfuss, A. Jahedi, Y . Nalivayko, and A. Bruhn, “Spring: A high-resolution high-detail dataset and benchmark for scene flow, optical flow and stereo,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 4981–4991. 6

    work page 2023

  70. [70]

    Large scale real-world multi-person tracking,

    B. Shuai, A. Bergamo, U. Buechler, A. Berneshawi, A. Boden, and J. Tighe, “Large scale real-world multi-person tracking,” inEuropean Conference on Computer Vision, 2022, pp. 504–521. 7

    work page 2022