Pith. sign in

REVIEW 4 major objections 6 minor 63 references

FusionTrack: End-to-End Multi-Object Tracking in Arbitrary Multi-View Environment

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read FusionTrack claims multi-view tracking can be one end-to-end Transformer model instead of two stages.

desk verdict A genuinely new drone-view MVMOT benchmark and a plausible end-to-end system, but the cross-dataset SOTA claims currently rest on unverifiable comparisons and a handful of internal inconsistencies. read the letter →

arxiv 2505.18727 v1 pith:BCGJNQBE submitted 2025-05-24 cs.CV

Xiaohe Li , Pengfei Li , Zide Fan , Ying Geng , Fangli Mou , Haohua Wu , Yunping Ge This is my paper · ORCID
classification cs.CV
keywords multi-viewmulti-objecttrackingarbitraryviewpointsdroneswarmend-to-endTransformerandre-identificationMDMOTdatasetcross-viewassociationhierarchicalclustering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

FusionTrack claims that multi-view multi-object tracking can be solved as one end-to-end Transformer problem rather than the usual two-stage sequence of single-view tracking followed by cross-view association. The paper introduces MDMOT, a drone-swarm dataset with overlapping and non-overlapping views, and a unified architecture that jointly optimizes detection, tracking, and re-identification. On MDMOT it reports CVMA 80.8 and CVIDF1 75.2, ahead of the two strongest prior pipelines, and it also reports top results on CAMPUS, WILDTRACK, MvMHAT, and DIVOTrack. If correct, this would make free-viewpoint cooperative tracking practical with a single trained model.

What carries the argument

The load-bearing object is the track query: a Transformer query token encoding one target's bounding box, class, and identity, carried across frames in a Tracklet Memory Pool and refined by an Object Update Module that first aggregates features from recent frames, then from views observing the same identity. A time-decayed cross-attention weight $W = e^{-\alpha T}$ implements temporal emphasis, and a view-positional encoding supplies camera context. Cross-view identity is resolved by hierarchical clustering on a cosine distance matrix masked so that two objects in the same view never merge, with a Neighbor Filtering Mechanism that keeps only matches whose spatial neighbors also agree.

What would settle it

Run FusionTrack on DIVOTrack and WILDTRACK using the exact detection boxes adopted by the published CrossMOT and MvMHAT baselines rather than the authors' own detections; if CVMA and CVIDF1 drop toward or below those baselines, the claimed state-of-the-art gain is not attributable to the architecture.

Watch

Extended reading notes

Core claim

The central claim is that fusing single-view tracking with cross-view re-identification inside one Transformer, instead of running them as separate stages, improves both tasks. The paper argues that the same object queries can carry appearance, motion, and identity information, updated across frames and across cameras by an Object Update Module, with a Tracklet Memory Pool preserving identities through occlusion and re-entry. At inference, view-aware masking, mutual top-k neighbor filtering, and hierarchical clustering assign global identities consistent with the cross-view geometry. The result is claimed to be state-of-the-art multi-view tracking on both the new MDMOT benchmark and four established pedestrian datasets.

Load-bearing premise

The reported comparison assumes that the published baseline numbers on CAMPUS, WILDTRACK, MvMHAT, and DIVOTrack were produced under the same detection inputs, resolution, and evaluation protocol as the authors' runs, but the paper gives implementation details only for MDMOT.

Editorial extensions

If this is right

  • A single network replaces the track-then-associate pipeline, removing error propagation between separately trained tracking and ReID stages.
  • Tracking under arbitrary numbers of cameras, including non-overlapping views, becomes one model with no per-scene camera calibration or homography estimation.
  • The MDMOT dataset gives the field a benchmark with dense urban scenes, moving cameras, and both overlapping and disjoint fields of view.
  • Joint single-view and multi-view training appears to lift single-view MOT metrics too, with FusionTrack reporting the best Transformer-based MOTA on MDMOT.
  • The reported gains over CrossMOT on DIVOTrack (2.3 CVMA, 6.1 CVIDF1) and over MvMHAT and CrossMOT on MDMOT suggest the architecture transfers across pedestrian and vehicle scenes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The cross-view neighbor consistency check assumes that physical proximity in one overlapping view predicts proximity in another; in non-overlapping views it cannot help, and robustness there would rest mainly on ReID features alone.
  • A natural extension would be to test the same architecture on datasets with more than five cameras or with very different object classes, since the query-update mechanism is class-agnostic.
  • The progressive training schedule (no cross-view matching for the first 20 epochs) suggests the method depends on a warm-stabilized tracker before ReID can help; a controlled study of when to enable cross-view updates could guide transfer to new domains.
  • If detection inputs were unified across all compared datasets, the reported margins might change; a fair re-benchmark with identical detections would separate architecture gains from detection gains.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper introduces FusionTrack, an end-to-end Transformer-based multi-view multi-object tracking framework that unifies single-view tracking and cross-view ReID, together with a new benchmark dataset MDMOT captured from drone swarms with both overlapping and non-overlapping views. The method uses a Tracklet Memory Pool, an Object Update Module, a Neighbor Filtering Mechanism, and optimal-transport-based global identity optimization. The authors report experiments on MDMOT and four existing pedestrian datasets (CAMPUS, WILDTRACK, MvMHAT, DIVOTrack), claiming state-of-the-art results in both single-view and multi-view tracking, and provide ablations and parameter studies.

Significance. The paper is potentially significant because MDMOT is, to the best of the authors' knowledge, the first drone-based multi-view tracking dataset with both overlapping and non-overlapping views and arbitrary viewpoints, which addresses a real gap in the MVMOT literature. The proposed FusionTrack architecture is a plausible end-to-end approach that jointly optimizes tracking and ReID, and the ablation study (Table V) shows that each proposed component contributes to the reported performance. The analysis of parameter sensitivity (Figs. 9-10) is a useful addition. However, the experimental validation is currently insufficient to support the claimed state of the art: the cross-dataset comparisons lack essential implementation details, and there are internal numerical inconsistencies that must be resolved.

major comments (4)
  1. [Section V.C.2 and Table III] The text reports CAMPUS CVMA of 66.8% while Table III reports 68.8%, and the text reports WILDTRACK CVIDF1 of 58.0% while Table III reports 60.8%. The text's claimed margin over CrossMOT on WILDTRACK CVIDF1 (4.1%) is consistent with the table value (60.8% - 56.7% = 4.1%), not with the text value (58.0%). These contradictions must be corrected, because the claimed SOTA margins are internally inconsistent.
  2. [Section V.A and V.C.2] Implementation details are provided only for the MDMOT dataset. For CAMPUS, WILDTRACK, MvMHAT, and DIVOTrack, the paper does not state which detections are used (ground-truth boxes, dataset-provided detections, or the model's own detector), whether the model is trained from scratch on each dataset, fine-tuned, or transferred from MDMOT, or what hyperparameters and training schedules are employed. Since multi-view tracking metrics depend critically on detection quality, the comparisons in Table III cannot be verified as apples-to-apples, and the reported gains over published baselines cannot be attributed to the FusionTrack architecture.
  3. [Abstract and Table II] The abstract and introduction claim state-of-the-art performance in both single-view and multi-view tracking, but Table II shows that OC-SORT (a CNN-based method) achieves higher MOTA (89.42 vs 88.13), MOTP (88.73 vs 87.92), IDF1 (92.15 vs 92.04), and HOTA (85.12 vs 84.42) than FusionTrack on MDMOT. The claim should be limited to 'best among Transformer-based methods,' as the text in Section V.C.1 correctly states.
  4. [Section III.B and Table I] The dataset statistics are contradictory: Section III.B states 'over 24,810,000 annotated bounding boxes,' whereas Table I lists 'Boxes 2481K' (i.e., 2,481,000). This factor-of-ten discrepancy must be resolved because it directly affects the characterization of the dataset's scale and the comparison with other datasets in Table I.
minor comments (6)
  1. [Table III] In the AGW row, the WILDTRACK and DIVOTrack columns are identical (15.6 and 23.8), which is likely a copy-paste error and should be checked.
  2. [Section IV.F.1] The role of the confidence threshold δ in the inference procedure is described informally; please specify how δ is used in the matching decision and whether it is applied before or after the mutual top-k step.
  3. [Section V.C.1] The sentence 'It surpasses most CNN-based methods' is imprecise, as OC-SORT outperforms FusionTrack on several metrics; please specify which methods are surpassed.
  4. [Section V] No error bars or multiple-run statistics are reported for any of the experiments. Given that some reported gains are small (e.g., Table IV, CVMA 80.8 vs 78.83 for MvMHAT), please report variance or state that results are from a single run.
  5. [Throughout] The manuscript contains numerous typos and formatting issues (e.g., 'P ipline' in Section IV.C, inconsistent reference formatting, and unexpanded abbreviations such as 'UA V' in Section V.C.3); a careful proofread is needed.
  6. [Section V.A] The paper does not mention whether the code and the MDMOT dataset will be released; for a benchmark contribution, a release plan is essential for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity. FusionTrack's SOTA claim is an empirical benchmark result measured against external baselines and human-annotated ground truth; the paper contains no self-citations and no fitted quantity is repackaged as a prediction.

full rationale

FusionTrack is an empirical systems paper: its central claims (state-of-the-art single-view and multi-view tracking; MDMOT dataset utility) are supported by measured numbers on MDMOT, CAMPUS, WILDTRACK, MvMHAT, and DIVOTrack against published external baselines (OSNet, Strong, AGW, CT, MGN, MvMHAT, CrossMOT). The evaluation metrics CVMA and CVIDF1 are explicitly adopted from the external CrossMOT work ([16]), and every architectural component is credited to external prior work (MOTR [36], TransReID [44], FairMOT [33]); scanning the reference list turned up no citations to the present authors' prior work, so no self-citation chain appears. Hyperparameters (memory window tau_1=30, ReID window tau_2=10, update window tau_3=6, threshold delta'=0.5) are selected on the authors' own MDMOT split (Section V.E) and then transferred to other datasets; this is standard tuning, not a fitted parameter renamed as a prediction, because the reported test numbers are evaluated against ground truth rather than constructed from the tuned values. The inference-time cross-view feature update (Section V.F.1) does use the model's own confidence-thresholded ReID matches (delta=0.8) to refine features before final clustering, which is an online self-refinement loop, but the final CVMA/CVIDF1 are measured against human-verified identities, so the prediction does not reduce to its input by construction. Flagged weaknesses are correctness/reproducibility risks, not circularity: Section V.A gives implementation details only for MDMOT and omits detection inputs and training protocols for CAMPUS/WILDTRACK/MvMHAT/DIVOTrack, making Table III's SOTA margins hard to audit; Section V.C.2's text (CAMPUS CVMA 66.8, WILDTRACK CVIDF1 58.0) conflicts with Table III (68.8, 60.8); and Section III.B's 'over 24,810,000 annotated bounding boxes' conflicts with Table I's 2481K. None of these undermine the conclusion that the reported results are externally measured rather than derived from the paper's own assumptions. The derivation chain is therefore self-contained and free of circular steps.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claims rest on standard deep learning background and on the correctness of the MDMOT annotations, which were produced semi-automatically with ByteTrack and manually refined. No new physical entities are introduced; TMP, OUM, and NFM are algorithmic modules.

free parameters (7)
  • Memory windows tau1, tau2, tau3 = 30, 10, 6
    Hand-chosen trajectory memory, ReID window, and cross-frame update window (Section V.A); parameter study only for tau3 (Fig. 9).
  • Temporal decay alpha = 0.5
    Fixed weight for time-decayed cross-frame attention (Section V.A).
  • ReID confidence threshold delta = 0.8
    Minimum cosine similarity to seed cross-view update during inference (Section V.A).
  • Top-k neighbor count = 10
    Mutual top-k candidates for cross-view matching (Section V.A).
  • NFM ratio threshold delta' = 0.5
    Neighbor consistency ratio for filtering matches, tuned on MDMOT (Fig. 10).
  • Loss weights lambda_cls, lambda_reg, lambda_giou = not specified
    Empirically determined (Section IV.D), no values reported.
  • Uncertainty weights initial w1, w2 = -1.85, -1.05
    Learnable task-balancing parameters initialized as in FairMOT (Section V.A).
assumptions (4)
  • domain assumption MDMOT annotations are correct and cross-view identity labels are consistent
    Labels were produced semi-automatically: ByteTrack pretrained on VisDrone plus manual DarkLabel refinement and manual cross-view association (Section III.A.2). If cross-view identities are wrong, training and evaluation on MDMOT are compromised.
  • domain assumption CVMA and CVIDF1 adequately measure multi-view tracking quality
    Metrics are adopted from DIVOTrack without independent validation (Section V.B).
  • domain assumption Pretrained DAB-Deformable-DETR weights transfer to drone imagery
    Backbone initialized with COCO-pretrained DAB-Deformable-DETR; no fine-tuning schedule analysis for domain gap (Section V.A).
  • domain assumption Cross-view evaluation on CAMPUS, WILDTRACK, MvMHAT, DIVOTrack follows official protocols
    No implementation details are given for these datasets, so comparability with reported baselines is assumed (Section V.C.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of FusionTrack: End-to-End Multi-Object Tracking in Arbitrary Multi-View Environment." pith.science (2026). https://pith.science/paper/BCGJNQBE

@misc{pith2026250518727,
  author       = {Pith},
  title        = {Pith review of: FusionTrack: End-to-End Multi-Object Tracking in Arbitrary Multi-View Environment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BCGJNQBE}},
  note         = {Machine review of arXiv:2505.18727}
}
read the original abstract

Multi-view multi-object tracking (MVMOT) has found widespread applications in intelligent transportation, surveillance systems, and urban management. However, existing studies rarely address genuinely free-viewpoint MVMOT systems, which could significantly enhance the flexibility and scalability of cooperative tracking systems. To bridge this gap, we first construct the Multi-Drone Multi-Object Tracking (MDMOT) dataset, captured by mobile drone swarms across diverse real-world scenarios, initially establishing the first benchmark for multi-object tracking in arbitrary multi-view environment. Building upon this foundation, we propose \textbf{FusionTrack}, an end-to-end framework that reasonably integrates tracking and re-identification to leverage multi-view information for robust trajectory association. Extensive experiments on our MDMOT and other benchmark datasets demonstrate that FusionTrack achieves state-of-the-art performance in both single-view and multi-view tracking.

Figures

Figures reproduced from arXiv: 2505.18727 by the authors.

Figure 1
Figure 1. Examples of the MDMOT dataset. The left panel illustrates the spatial distribution of drones at a specific moment within [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Image examples under varying weather conditions. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) Heatmap of object location distribution, where the x [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Comparison of dataset statistics, with blue indicating [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Overview of our FusionTrack framework. It comprises the Single-view Tracking module, which tracks objects [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Pipline of the Object Update Module. We update the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Post-processing pipeline, consisting of view-based masking, top-k masking, Neighbor Filtering, and Viewpoint-guided [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Implementation of the NFM. Dashed lines indicate [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Variation of CVIDF1 with the change in cross-frame [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Variation of CVIDF1 with the change in NFM [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Illustrative results of associated tracking in intersection scene. [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Illustrative results of associated tracking in overpass scene. [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 51 canonical work pages

  1. [1]

    Multiple object tracking in recent times: A literature review

    Mk Bashar, Samia Islam, Kashifa Kawaakib Hussain, Md Bakhtiar Hasan, ABM Rahman, and Md Hasanul Kabir. Multiple object tracking in recent times: A literature review. arXiv preprint arXiv:2209.04796 , 2022

    arXiv 2022

  2. [2]

    Robust superpixel tracking

    Fan Yang, Huchuan Lu, and Ming-Hsuan Yang. Robust superpixel tracking. IEEE Transactions on Image Processing , 23(4):1639–1651, 2014

    work page 2014

  3. [3]

    Gmt: A robust global association model for multi-target multi-camera tracking

    Huijie Fan, Tinghui Zhao, Qiang Wang, Baojie Fan, Yandong Tang, and LianQing Liu. Gmt: A robust global association model for multi-target multi-camera tracking. arXiv preprint arXiv:2407.01007 , 2024

    arXiv 2024

  4. [4]

    Video-based multi-camera vehicle tracking via appearance- parsing spatio-temporal trajectory matching network

    Xiaoqin Zhang, Hongqi Yu, Yong Qin, Xiaolong Zhou, and Sixian Chan. Video-based multi-camera vehicle tracking via appearance- parsing spatio-temporal trajectory matching network. IEEE Transactions on Circuits and Systems for Video Technology , 2024

    work page 2024

  5. [5]

    Drone-based rgb- infrared cross-modality vehicle detection via uncertainty-aware learning

    Yiming Sun, Bing Cao, Pengfei Zhu, and Qinghua Hu. Drone-based rgb- infrared cross-modality vehicle detection via uncertainty-aware learning. IEEE Transactions on Circuits and Systems for Video Technology , 32(10):6700–6713, 2022

    work page 2022

  6. [6]

    Bayesian loss for crowd count estimation with point supervision

    Zhiheng Ma, Xing Wei, Xiaopeng Hong, and Yihong Gong. Bayesian loss for crowd count estimation with point supervision. In Proceedings of the IEEE/CVF international conference on computer vision , pages 6142–6151, 2019

    work page 2019

  7. [7]

    Multiple pedestrian tracking under occlusion: A survey and outlook

    Zhihong Sun, Guoheng Wei, Wei Fu, Mang Ye, Kui Jiang, Chao Liang, Tingting Zhu, Tao He, and Mithun Mukherjee. Multiple pedestrian tracking under occlusion: A survey and outlook. IEEE Transactions on Circuits and Systems for Video Technology , 2024

    work page 2024

  8. [8]

    Cityflow: A city-scale benchmark for multi-target multi- camera vehicle tracking and re-identification

    Zheng Tang, Milind Naphade, Ming-Yu Liu, Xiaodong Yang, Stan Birchfield, Shuo Wang, Ratnesh Kumar, David Anastasiu, and Jenq- Neng Hwang. Cityflow: A city-scale benchmark for multi-target multi- camera vehicle tracking and re-identification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8797–8806, 2019

    work page 2019

Show all 63 references
  1. [9]

    Robust multi-drone multi-target tracking to resolve target occlusion: A benchmark

    Zhihao Liu, Yuanyuan Shang, Timing Li, Guanlin Chen, Yu Wang, Qinghua Hu, and Pengfei Zhu. Robust multi-drone multi-target tracking to resolve target occlusion: A benchmark. IEEE Transactions on Multimedia, 25:1462–1476, 2023

    2023

  2. [10]

    Multi-camera multi-object tracking: A review of current trends and future advances

    Temitope Ibrahim Amosa, Patrick Sebastian, Lila Iznita Izhar, Oladimeji Ibrahim, Lukman Shehu Ayinla, Abdulrahman Abdullah Bahashwan, Abubakar Bala, and Yau Alhaji Samaila. Multi-camera multi-object tracking: A review of current trends and future advances. Neurocom- puting, 55...

    2023

  3. [11]

    Integrating social grouping for multitarget tracking across cameras in a crf model

    Xiaojing Chen and Bir Bhanu. Integrating social grouping for multitarget tracking across cameras in a crf model. IEEE Transactions on Circuits and Systems for Video Technology , 27(11):2382–2394, 2016

    2016

  4. [12]

    Multicamera people tracking with a probabilistic occupancy map

    Francois Fleuret, Jerome Berclaz, Richard Lengagne, and Pascal Fua. Multicamera people tracking with a probabilistic occupancy map. IEEE transactions on pattern analysis and machine intelligence , 30(2):267– 282, 2007

    2007

  5. [13]

    Multi-view people tracking via hierarchical trajectory composition

    Yuanlu Xu, Xiaobai Liu, Yang Liu, and Song-Chun Zhu. Multi-view people tracking via hierarchical trajectory composition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 4256–4265, 2016

    2016

  6. [14]

    Self-supervised multi-view multi-human association and tracking

    Yiyang Gan, Ruize Han, Liqiang Yin, Wei Feng, and Song Wang. Self-supervised multi-view multi-human association and tracking. In Proceedings of the 29th ACM international conference on multimedia , pages 282–290, 2021

    2021

  7. [15]

    Mmptrack: Large- scale densely annotated multi-camera multiple people tracking bench- mark

    Xiaotian Han, Quanzeng You, Chunyu Wang, Zhizheng Zhang, Peng Chu, Houdong Hu, Jiang Wang, and Zicheng Liu. Mmptrack: Large- scale densely annotated multi-camera multiple people tracking bench- mark. In Proceedings of the IEEE/CVF Winter Conference on Applica- tions of Compute...

    2023

  8. [16]

    Divotrack: A novel dataset and baseline method for cross-view multi-object tracking in diverse open scenes

    Shengyu Hao, Peiyuan Liu, Yibing Zhan, Kaixun Jin, Zuozhu Liu, Mingli Song, Jenq-Neng Hwang, and Gaoang Wang. Divotrack: A novel dataset and baseline method for cross-view multi-object tracking in diverse open scenes. International Journal of Computer Vision , 132(4):1075–1090, 2024

    2024

  9. [17]

    Bucktales: A multi-uav dataset for multi-object tracking and re-identification of wild antelopes

    Hemal Naik, Junran Yang, Dipin Das, Margaret Crofoot, Akanksha Rathore, and Vivek Hari Sridhar. Bucktales: A multi-uav dataset for multi-object tracking and re-identification of wild antelopes. Advances in Neural Information Processing Systems , 37:81992–82009, 2024

    2024

  10. [18]

    Mitracker: Multi-view integration for visual object tracking

    Mengjie Xu, Yitao Zhu, Haotian Jiang, Jiaming Li, Zhenrong Shen, Sheng Wang, Haolin Huang, Xinyu Wang, Qing Yang, Han Zhang, et al. Mitracker: Multi-view integration for visual object tracking. arXiv preprint arXiv:2502.20111, 2025

    2025 arXiv

  11. [19]

    Simple online and realtime tracking

    Alex Bewley, Zongyuan Ge, Lionel Ott, Fabio Ramos, and Ben Upcroft. Simple online and realtime tracking. In 2016 IEEE international conference on image processing (ICIP) , pages 3464–3468. Ieee, 2016

    2016

  12. [20]

    Simple online and realtime tracking with a deep association metric

    Nicolai Wojke, Alex Bewley, and Dietrich Paulus. Simple online and realtime tracking with a deep association metric. In 2017 IEEE international conference on image processing (ICIP), pages 3645–3649. IEEE, 2017

    2017

  13. [21]

    Bytetrack: Multi- object tracking by associating every detection box

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. Bytetrack: Multi- object tracking by associating every detection box. In European conference on computer vision , pages 1–21. Springer, 2022

    2022

  14. [22]

    Distinctive image features from scale-invariant keypoints

    David G Low. Distinctive image features from scale-invariant keypoints. Journal of Computer Vision , 60(2):91–110, 2004

    2004

  15. [23]

    Multi-target multi-camera tracking by tracklet-to-target assign- ment

    Yuhang He, Xing Wei, Xiaopeng Hong, Weiwei Shi, and Yihong Gong. Multi-target multi-camera tracking by tracklet-to-target assign- ment. IEEE Transactions on Image Processing , 29:5191–5205, 2020

    2020

  16. [24]

    Lammon: language model combined graph neural network for multi- target multi-camera tracking in online scenarios

    Tuan T Nguyen, Hoang H Nguyen, Mina Sartipi, and Marco Fisichella. Lammon: language model combined graph neural network for multi- target multi-camera tracking in online scenarios. Machine Learning , 113(9):6811–6837, 2024

    2024

  17. [25]

    Open world object detection: a survey

    Yiming Li, Yi Wang, Wenqian Wang, Dan Lin, Bingbing Li, and Kim- Hui Yap. Open world object detection: a survey. IEEE Transactions on Circuits and Systems for Video Technology , 2024

    2024

  18. [26]

    Fast r-cnn

    Ross Girshick. Fast r-cnn. In Proceedings of the IEEE international conference on computer vision , pages 1440–1448, 2015

    2015

  19. [27]

    Cascade r-cnn: Delving into high quality object detection

    Zhaowei Cai and Nuno Vasconcelos. Cascade r-cnn: Delving into high quality object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 6154–6162, 2018

    2018

  20. [28]

    A review of yolo algorithm developments

    Peiyuan Jiang, Daji Ergu, Fangyao Liu, Ying Cai, and Bo Ma. A review of yolo algorithm developments. Procedia computer science, 199:1066– 1073, 2022

    2022

  21. [29]

    Ssd: Single shot multibox detector

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14, p...

    2016

  22. [30]

    Ctif-net: A cnn-transformer iterative fusion network for salient object detection

    Junbin Yuan, Aiqing Zhu, Qingzhen Xu, Kanoksak Wattanachote, and Yongyi Gong. Ctif-net: A cnn-transformer iterative fusion network for salient object detection. IEEE Transactions on Circuits and Systems for Video Technology, 34(5):3795–3805, 2023

    2023

  23. [31]

    Tracktor: image-based automated tracking of animal movement and behaviour

    Vivek Hari Sridhar, Dominique G Roche, and Simon Gingins. Tracktor: image-based automated tracking of animal movement and behaviour. Methods in Ecology and Evolution , 10(6):815–820, 2019

    2019

  24. [32]

    Towards real-time multi-object tracking

    Zhongdao Wang, Liang Zheng, Yixuan Liu, Yali Li, and Shengjin Wang. Towards real-time multi-object tracking. In European conference on computer vision, pages 107–122. Springer, 2020

    2020

  25. [33]

    Fairmot: On the fairness of detection and re-identification in multiple object tracking

    Yifu Zhang, Chunyu Wang, Xinggang Wang, Wenjun Zeng, and Wenyu Liu. Fairmot: On the fairness of detection and re-identification in multiple object tracking. International journal of computer vision , 129:3069–3087, 2021

    2021

  26. [34]

    Transtrack: Multiple object tracking with transformer

    Peize Sun, Jinkun Cao, Yi Jiang, Rufeng Zhang, Enze Xie, Zehuan Yuan, Changhu Wang, and Ping Luo. Transtrack: Multiple object tracking with transformer. arXiv preprint arXiv:2012.15460 , 2020

    2012 arXiv

  27. [35]

    Trackformer: Multi-object tracking with transformers

    Tim Meinhardt, Alexander Kirillov, Laura Leal-Taixe, and Christoph Feichtenhofer. Trackformer: Multi-object tracking with transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8844–8854, 2022

    2022

  28. [36]

    Motr: End-to-end multiple-object tracking with JOURNAL OF LATEX CLASS FILES, VOL

    Fangao Zeng, Bin Dong, Yuang Zhang, Tiancai Wang, Xiangyu Zhang, and Yichen Wei. Motr: End-to-end multiple-object tracking with JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 15 transformer. In European conference on computer vision , pages 659–

    2021

  29. [37]

    Memotr: Long-term memory-augmented transformer for multi-object tracking

    Ruopeng Gao and Limin Wang. Memotr: Long-term memory-augmented transformer for multi-object tracking. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 9901–9910, 2023

    2023

  30. [38]

    Mambavt: Spatio-temporal contextual modeling for robust rgb-t tracking

    Simiao Lai, Chang Liu, Jiawen Zhu, Ben Kang, Yang Liu, Dong Wang, and Huchuan Lu. Mambavt: Spatio-temporal contextual modeling for robust rgb-t tracking. arXiv preprint arXiv:2408.07889 , 2024

    2024 arXiv

  31. [39]

    Mask-aware hierarchical aggregation transformer for occluded person re-identification

    Guoqing Zhang, Yan Yang, Yuhui Zheng, Gaven Martin, and Ruili Wang. Mask-aware hierarchical aggregation transformer for occluded person re-identification. IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  32. [40]

    Beyond part models: Person retrieval with refined part pooling (and a strong convolutional baseline)

    Yifan Sun, Liang Zheng, Yi Yang, Qi Tian, and Shengjin Wang. Beyond part models: Person retrieval with refined part pooling (and a strong convolutional baseline). In Proceedings of the European conference on computer vision (ECCV) , pages 480–496, 2018

    2018

  33. [41]

    Learning discriminative features with multiple granularities for person re-identification

    Guanshuo Wang, Yufeng Yuan, Xiong Chen, Jiwei Li, and Xi Zhou. Learning discriminative features with multiple granularities for person re-identification. In Proceedings of the 26th ACM international confer- ence on Multimedia , pages 274–282, 2018

    2018

  34. [42]

    Semantics-aligned representation learning for person re-identification

    Xin Jin, Cuiling Lan, Wenjun Zeng, Guoqiang Wei, and Zhibo Chen. Semantics-aligned representation learning for person re-identification. In Proceedings of the AAAI Conference on Artificial Intelligence , vol- ume 34, pages 11173–11180, 2020

    2020

  35. [43]

    Deep learning for person re-identification: A survey and outlook

    Mang Ye, Jianbing Shen, Gaojie Lin, Tao Xiang, Ling Shao, and Steven CH Hoi. Deep learning for person re-identification: A survey and outlook. IEEE transactions on pattern analysis and machine intelligence, 44(6):2872–2893, 2021

    2021

  36. [44]

    Transreid: Transformer-based object re-identification

    Shuting He, Hao Luo, Pichao Wang, Fan Wang, Hao Li, and Wei Jiang. Transreid: Transformer-based object re-identification. In Proceedings of the IEEE/CVF international conference on computer vision , pages 15013–15022, 2021

    2021

  37. [45]

    Hat: Hierarchical aggregation transformers for person re-identification

    Guowen Zhang, Pingping Zhang, Jinqing Qi, and Huchuan Lu. Hat: Hierarchical aggregation transformers for person re-identification. In Proceedings of the 29th ACM international conference on multimedia , pages 516–525, 2021

    2021

  38. [46]

    Multi-target multi-camera tracking with optical-based pose association

    Sisi You, Hantao Yao, and Changsheng Xu. Multi-target multi-camera tracking with optical-based pose association. IEEE Transactions on Circuits and Systems for Video Technology , 31(8):3105–3117, 2020

    2020

  39. [47]

    Spatio- temporal consistency and hierarchical matching for multi-target multi- camera vehicle tracking

    Peilun Li, Guozhen Li, Zhangxi Yan, Youzeng Li, Meiqi Lu, Pengfei Xu, Yang Gu, Bing Bai, Yifei Zhang, and DiDi Chuxing. Spatio- temporal consistency and hierarchical matching for multi-target multi- camera vehicle tracking. In CVPR Workshops, pages 222–230, 2019

    2019

  40. [48]

    An equalized global graph model-based approach for multicamera object tracking

    Weihua Chen, Lijun Cao, Xiaotang Chen, and Kaiqi Huang. An equalized global graph model-based approach for multicamera object tracking. IEEE Transactions on Circuits and Systems for Video Tech- nology, 27(11):2367–2381, 2016

    2016

  41. [49]

    Hypergraphs for joint multi-view reconstruction and multi-object tracking

    Martin Hofmann, Daniel Wolf, and Gerhard Rigoll. Hypergraphs for joint multi-view reconstruction and multi-object tracking. In Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3650–3657, 2013

    2013

  42. [50]

    Dyglip: A dynamic graph model with link prediction for accurate multi-camera multiple object tracking

    Kha Gia Quach, Pha Nguyen, Huu Le, Thanh-Dat Truong, Chi Nhan Duong, Minh-Triet Tran, and Khoa Luu. Dyglip: A dynamic graph model with link prediction for accurate multi-camera multiple object tracking. In Proceedings of the IEEE/CVF conference on computer vision and pattern r...

    2021

  43. [51]

    Box-grained reranking matching for multi-camera multi-target tracking

    Xipeng Yang, Jin Ye, Jincheng Lu, Chenting Gong, Minyue Jiang, Xiangru Lin, Wei Zhang, Xiao Tan, Yingying Li, Xiaoqing Ye, et al. Box-grained reranking matching for multi-camera multi-target tracking. In Proceedings of the IEEE/CVF conference on computer vision and pattern rec...

    2022

  44. [52]

    Multi-camera multi- object tracking

    Wenqian Liu, Octavia Camps, and Mario Sznaier. Multi-camera multi- object tracking. arXiv preprint arXiv:1709.07065 , 2017

    2017 arXiv

  45. [53]

    The wildtrack multi-camera person dataset

    Tatjana Chavdarova, Pierre Baqu ´e, St ´ephane Bouquet, Andrii Maksai, Cijo Jose, Louis Lettry, Pascal Fua, Luc Van Gool, and Franc ¸ois Fleuret. The wildtrack multi-camera person dataset. arXiv preprint arXiv:1707.09299, 2017

    2017 arXiv

  46. [54]

    Visdrone- mot2019: The vision meets drone multiple object tracking challenge results

    Longyin Wen, Pengfei Zhu, Dawei Du, Xiao Bian, Haibin Ling, Qinghua Hu, Jiayu Zheng, Tao Peng, Xinyao Wang, Yue Zhang, et al. Visdrone- mot2019: The vision meets drone multiple object tracking challenge results. In Proceedings of the IEEE/CVF International Conference on Comput...

    2019

  47. [55]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 770–778, 2016

    2016

  48. [56]

    Dab-detr: Dynamic anchor boxes are better queries for detr

    S Liu, F Li, H Zhang, X Yang, X Qi, H Su, J Zhu, and L Zhang. Dab-detr: Dynamic anchor boxes are better queries for detr. arxiv 2022. arXiv preprint arXiv:2201.12329

    2022 arXiv

  49. [57]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer vision–ECCV 2014: 13th European conference, zurich, Switzerland, September 6-12, 2014, proceeding...

    2014

  50. [58]

    Tracking objects as points

    Xingyi Zhou, Vladlen Koltun, and Philipp Kr ¨ahenb¨uhl. Tracking objects as points. In European conference on computer vision , pages 474–490. Springer, 2020

    2020

  51. [59]

    Track to detect and segment: An online multi-object tracker

    Jialian Wu, Jiale Cao, Liangchen Song, Yu Wang, Ming Yang, and Junsong Yuan. Track to detect and segment: An online multi-object tracker. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 12352–12361, 2021

    2021

  52. [60]

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

    Jinkun Cao, Jiangmiao Pang, Xinshuo Weng, Rawal Khirodkar, and Kris Kitani. Observation-centric sort: Rethinking sort for robust multi-object tracking. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 9686–9696, 2023

    2023

  53. [61]

    Omni- scale feature learning for person re-identification

    Kaiyang Zhou, Yongxin Yang, Andrea Cavallaro, and Tao Xiang. Omni- scale feature learning for person re-identification. In Proceedings of the IEEE/CVF international conference on computer vision , pages 3702– 3712, 2019

    2019

  54. [62]

    Bag of tricks and a strong baseline for deep person re-identification

    Hao Luo, Youzhi Gu, Xingyu Liao, Shenqi Lai, and Wei Jiang. Bag of tricks and a strong baseline for deep person re-identification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops , pages 0–0, 2019

    2019

  55. [63]

    On the unreasonable effectiveness of centroids in image retrieval

    Mikolaj Wieczorek, Barbara Rychalska, and Jacek Dkabrowski. On the unreasonable effectiveness of centroids in image retrieval. In Neural Information Processing: 28th International Conference, ICONIP 2021, Sanur, Bali, Indonesia, December 8–12, 2021, Proceedings, Part IV 28 , p...

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.