Pith. sign in

REVIEW 3 major objections 6 minor 59 references

CrowdTrack: A Benchmark for Difficult Multiple Pedestrian Tracking in Real Scenarios

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read CrowdTrack, a new benchmark of 33 dense first-person videos, makes every state-of-the-art pedestrian tracker it tests perform worse than on existing MOT benchmarks.

desk verdict CrowdTrack is a useful new benchmark, but the missing annotation-quality measurement makes the ID Switch results hard to interpret. read the letter →

arxiv 2507.02479 v1 pith:FNC2NJ77 submitted 2025-07-03 cs.CV cs.AI

Teng Fu , Yuwen Chen , Zhuofan Chen , Mengyang Zhao , Bin Li , Xiangyang Xue This is my paper
classification cs.CVcs.AI
keywords multi-objecttrackingpedestrianbenchmarkdatasetfirst-personviewcrowdedscenestracking-by-detectionHOTAmotionanalysis
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

The paper tries to establish that real-world pedestrian tracking is harder than current benchmarks suggest, and that the field needs a dataset reflecting dense, first-person, blurry, and heavily occluded scenes. It presents CrowdTrack, consisting of 33 videos, 5,185 trajectories, and over 700,000 person annotations, and shows that seven state-of-the-art trackers all lose accuracy on it relative to MOT17 and DanceTrack. If the paper is right, progress measured on existing benchmarks may overstate how ready current trackers are for everyday crowded environments like malls, stations, and streets.

What carries the argument

The load-bearing object is the dataset itself: 33 real-life videos, mostly first-person, with every visible person assigned a full-body bounding box and a unique track ID that persists through temporary disappearances, split into 17 training and 16 test videos. The paper characterizes the dataset's difficulty with four quantitative measures—mean inter-frame IoU, relative position-switch frequency, direction-change angle, and overlap-based crowdedness (Real IoU)—plus an appearance-similarity score. What carries the argument is the combination of these metrics placing CrowdTrack at the hard end of existing datasets, together with a benchmark protocol that reports MOTA and HOTA (higher-order tracking accuracy, balancing detection and association) for the same methods so the degradation is directly comparable.

What would settle it

Have two independent annotation teams label the same subset of the most crowded CrowdTrack videos and compute agreement on identity continuity and box overlap; low agreement in the densest clips would mean the reported ID Switch gaps between trackers could reflect annotation noise rather than tracker failure.

Watch

Extended reading notes

Core claim

The paper's central claim is that existing MOT benchmarks under-represent the difficulty of real crowded first-person scenes, and it introduces CrowdTrack to close that gap. On CrowdTrack, every evaluated tracker with comparable numbers—FairMOT, TraDeS, MeMOTR, DiffusionTrack, TrackFormer, OC-SORT, and ByteTrack—scores lower in MOTA and HOTA than on MOT17 or DanceTrack, and DiffusionTrack's ID Switch rises to 2–4 times that of the other methods. The paper also argues that detection, not just association, is a major bottleneck, because MOTA drops more sharply than HOTA. Additionally, it probes foundation-model features and finds that light training improves separability but pedestrians remain confusable due to high visual similarity.

Load-bearing premise

The conclusions rest on the assumption that the paid annotators correctly kept each person's label continuous and drew accurate boxes even in the most crowded, blurred clips, yet the paper reports no independent check on label consistency.

Editorial extensions

If this is right

  • Models that lead on MOT17 or DanceTrack cannot be assumed ready for dense first-person scenes; the benchmark provides concrete evidence that score order and identity stability change under crowd realism.
  • Because MOTA drops more than HOTA, the bottleneck is partly detection: small, partially visible, and blurred persons are missed, so detection improvement is a necessary condition for better tracking in such scenes.
  • DiffusionTrack's 2–4x higher ID Switch on CrowdTrack shows that generative or end-to-end trackers that look competitive on standard benchmarks can be especially fragile at identity association in dense crowds.
  • When ground-truth detections are used and IoU alone is the distance metric, HOTA is highest, suggesting that spatial overlap is the most reliable cue in these scenes and that appearance and motion models need to be rethought for this setting.

Reading between the lines

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

  • If annotation quality holds up, CrowdTrack could be reused as a stress test for embodied agents such as robots or phones navigating malls and stations, since the first-person viewpoint couples ego-motion with object motion in ways static-camera benchmarks do not.
  • The difficulty metrics are separable, so a natural follow-up is to build controlled subsets that vary one axis at a time, isolating which factor—density, blur, ego-motion, or appearance similarity—does the most damage to each tracker family.
  • The appearance-similarity result suggests that identity cues in everyday crowds are genuinely weaker, yielding a testable prediction: training a ReID model on CrowdTrack will improve generalization to other crowded datasets, while training on existing benchmarks will not.
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

3 major / 6 minor

Summary. The paper introduces CrowdTrack, a new multi-pedestrian tracking benchmark comprising 33 real-world video sequences, 5,185 trajectories, and approximately 703K bounding-box annotations, mostly from first-person views in dense, occluded, and blurred scenes. The authors argue that existing MOT datasets are too simple or unrealistic, and they support this by (i) analyzing the dataset with motion, crowdedness, and appearance similarity metrics; (ii) benchmarking eight SOTA trackers (FairMOT, TraDeS, MeMOTR, DiffusionTrack, TrackFormer, OC-SORT, ByteTrack) and reporting that all degrade on CrowdTrack relative to MOT17 and DanceTrack; (iii) isolating the effect of motion, appearance, and IoU cues under ground-truth detections; and (iv) providing a short exploratory analysis of foundation models (BLIP2, CLIP) on the data. The dataset and project code are released.

Significance. If the central claims hold, CrowdTrack is a valuable resource: it is larger than MOT17 in trajectories and annotations, is drawn from natural first-person scenarios, and provides a stress test that current trackers fail. The paper explicitly releases the dataset and code, which supports reproducibility, and the dataset statistics (IoU, relative-position switch, direction-change angle, crowdedness) are useful beyond the benchmark results. The finding that DiffusionTrack's ID Switch rises to 2-4x that of other methods is a falsifiable benchmark observation worth checking. The main significance gap is that the benchmark's credibility rests on annotation quality that is asserted but not measured; without quantitative annotation reliability, the headline 'all trackers degrade' result could be an artifact of label noise in precisely the hardest annotation cases (occlusion, identity continuity, blur).

major comments (3)
  1. [§3.1, §4.2, Table 2] The paper does not report any inter-annotator agreement, identity-consistency validation, or per-video QA numbers for the commercial annotation process. In dense, blurred first-person footage, the decisions most likely to be noisy are exactly those that the text describes: full-body boxes for partially occluded persons, exclusion of fully occluded persons, and ID continuity through temporary disappearances. Since the central claim of Table 2—all SOTA trackers degrade on CrowdTrack, with DiffusionTrack's ID.Sw 2-4x higher—depends on ground-truth track identities and boxes, the absence of annotation-quality measurement leaves open the possibility that the observed degradation, and in particular the ID.Sw outlier, reflects label noise rather than tracker behavior. Please add a double-annotated subset with agreement metrics (e.g., box IoU, ID continuity agreement, missed-object rate), report per-video annotation QA statistics, and discuss how annotation uncertainty propagates to the benchmark conclusions.
  2. [§4.1, §4.2, Table 2] Detector accuracy is not controlled when comparing trackers across datasets. The text states that default configurations are used for all methods, but for TBD methods the detector is part of the pipeline, and for MOT17 and DanceTrack the numbers appear to be taken from the original papers (indicated by '-' for unreported metrics). As a result, the claim in §4.2 that 'MOTA declines more significantly than HOTA, indicating that object detection is more challenging' is confounded by differences in detector architecture, training data, and domain shift across datasets. Please either run all TBD methods on a shared public detection set for CrowdTrack, or report detector-only metrics (e.g., AP, recall) separately and analyze how detection performance changes across datasets before attributing the MOTA drop to the dataset's detection difficulty.
  3. [§3.1, §3.2] The 'empirical size threshold' for excluding small objects is never specified. This threshold directly determines which persons are annotated, and therefore affects trajectory counts, density statistics, and every downstream benchmark number. The paper should report the threshold (e.g., in pixels, relative to image height, or by instance area) and provide a sensitivity analysis showing how the dataset statistics and benchmark rankings change with reasonable threshold variations. Additionally, Section 3.2 states that 'all pedestrians labeled' is a design principle, which is in tension with the threshold-based exclusion in Section 3.1; this inconsistency should be resolved.
minor comments (6)
  1. [§3.2, Eq. (3)] Equation (3) contains broken vector notation ('N(− − − − − →Bt iBt+τ i )') that makes the formula unreadable; please fix the typesetting.
  2. [§3.2, Eq. (4)] Equation (4) has an unmatched parenthesis in the definition of SRIoU; please correct it.
  3. [§5] The conclusion contains a grammatical error: 'We introduces CrowdTrack' should be 'We introduce CrowdTrack.'
  4. [§2.1] The method name 'Transtrack' should be capitalized consistently as 'TransTrack' (as in the reference).
  5. [§2.2, References [38]] The text refers to 'Youtube-VIS' but cites the YouTube-VOS paper; please correct the dataset name or citation.
  6. [§4.4] The foundation-model analysis is exploratory and appears in a 'Discussion' section; consider moving it to the supplementary material or clearly marking it as preliminary, since it does not constitute a benchmark result and is not integrated into the main conclusions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CrowdTrack's benchmark claims are measured dataset statistics and external tracker evaluations, not derivations from fitted inputs or self-citations.

full rationale

CrowdTrack is primarily a dataset-construction and benchmarking paper. Its headline numbers (33 videos, 5,185 trajectories, 703k annotations, and the per-method MOTA/HOTA/ID.Sw values in Table 2) are quantities measured on a newly collected test split using standard external evaluation tooling; they are not obtained by fitting parameters that were themselves defined from the target values. The dataset statistics in Eqs. (1)-(5) (IoU score, relative position switching, direction angle, crowdedness, appearance similarity) are descriptive summaries of the annotations, and none of them is used as an input to a prediction that is then reported as a confirmation. The small-object exclusion threshold and the manual scene split are design choices, not fitted parameters, and they do not by construction determine the observed degradation or the DiffusionTrack ID.Sw outlier. The only self-citations are [3] and [11], used in related-work and motivation contexts; neither is load-bearing for the dataset claim or for the benchmark conclusions. The principal weakness--no inter-annotator agreement or identity-consistency validation reported for the commercial labeling process--is a measurement-validity concern about ground-truth quality, not a circularity of the kind where an output is equivalent to an input by definition. Therefore the paper is self-contained against external benchmarks and receives a circularity score of 0.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

No new physical or model entities are postulated. The ledger consists of dataset design choices: the unstated small-object exclusion threshold can change trajectory counts and therefore every benchmark number, and the manual train/test split determines the reported results. The biggest unverified premise is annotation quality in crowded blurred video.

free parameters (2)
  • empirical size threshold for omitting small objects = not reported
    Section 3.1 states that 'objects below an empirical size threshold are excluded'. The threshold is chosen by hand and is not specified numerically, yet it directly determines which pedestrians are in the benchmark.
  • train/test scene split weights = 17 train / 16 test videos
    The paper says scenes were split to evenly distribute videos, tracklets, and annotations. The split is a manual design choice that affects all reported benchmark numbers.
assumptions (3)
  • domain assumption The commercial annotations are accurate enough that track ID errors are negligible for benchmark conclusions.
    Invoked throughout Section 3.1 and Section 4.2. No inter-annotator agreement or manual verification rate is reported, and identity consistency in long dense sequences is assumed for all downstream metrics.
  • domain assumption Standard MOT metrics (MOTA, HOTA, IDF1) computed on CrowdTrack are comparable to the same metrics computed on MOT17 and DanceTrack.
    Invoked in Section 4.2. The datasets differ in frame rate, camera motion, density, sequence length, and annotation protocol, so any comparison of absolute scores assumes metric comparability across benchmarks.
  • domain assumption Using default model configurations and pretrained weights without fine-tuning on CrowdTrack measures generalization rather than poor hyperparameter choice.
    Invoked in Section 4.1. The authors state that default configurations are used and kept identical. If defaults are suboptimal for first-person dense scenes, the performance drop is partly due to configuration, not algorithmic weakness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CrowdTrack: A Benchmark for Difficult Multiple Pedestrian Tracking in Real Scenarios." pith.science (2026). https://pith.science/paper/FNC2NJ77

@misc{pith2026250702479,
  author       = {Pith},
  title        = {Pith review of: CrowdTrack: A Benchmark for Difficult Multiple Pedestrian Tracking in Real Scenarios},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FNC2NJ77}},
  note         = {Machine review of arXiv:2507.02479}
}
read the original abstract

Multi-object tracking is a classic field in computer vision. Among them, pedestrian tracking has extremely high application value and has become the most popular research category. Existing methods mainly use motion or appearance information for tracking, which is often difficult in complex scenarios. For the motion information, mutual occlusions between objects often prevent updating of the motion state; for the appearance information, non-robust results are often obtained due to reasons such as only partial visibility of the object or blurred images. Although learning how to perform tracking in these situations from the annotated data is the simplest solution, the existing MOT dataset fails to satisfy this solution. Existing methods mainly have two drawbacks: relatively simple scene composition and non-realistic scenarios. Although some of the video sequences in existing dataset do not have the above-mentioned drawbacks, the number is far from adequate for research purposes. To this end, we propose a difficult large-scale dataset for multi-pedestrian tracking, shot mainly from the first-person view and all from real-life complex scenarios. We name it ``CrowdTrack'' because there are numerous objects in most of the sequences. Our dataset consists of 33 videos, containing a total of 5,185 trajectories. Each object is annotated with a complete bounding box and a unique object ID. The dataset will provide a platform to facilitate the development of algorithms that remain effective in complex situations. We analyzed the dataset comprehensively and tested multiple SOTA models on our dataset. Besides, we analyzed the performance of the foundation models on our dataset. The dataset and project code is released at: https://github.com/loseevaya/CrowdTrack .

Figures

Figures reproduced from arXiv: 2507.02479 by the authors.

Figure 1
Figure 1. Comparison of our dataset with existing datasets. The images in (a) are from MOT17 [ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Some sampled scenes from the proposed dataset. All of our data comes from real life, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Some quantitative analyses of the proposed dataset. (a) compares the average overlap of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Average similarity of validation set sequences in each dataset.The green, blue and orange [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Some visualization from foundation models on our datasets. (a) shows the similarity [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 31 canonical work pages

  1. [1]

    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

    work page 2016

  2. [2]

    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

  3. [3]

    Foundation model driven appearance extraction for robust multiple object tracking

    Teng Fu, Haiyang Yu, Ke Niu, Bin Li, and Xiangyang Xue. Foundation model driven appearance extraction for robust multiple object tracking. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 3031–3039, 2025

    work page 2025

  4. [4]

    Memot: Multi-object tracking with memory

    Jiarui Cai, Mingze Xu, Wei Li, Yuanjun Xiong, Wei Xia, Zhuowen Tu, and Stefano Soatto. Memot: Multi-object tracking with memory. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8090–8100, 2022

    work page 2022

  5. [5]

    Track- former: Multi-object tracking with transformers

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

    work page 2022

  6. [6]

    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

    work page 2022

  7. [7]

    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

    work page 2021

  8. [8]

    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

    work page 2023

Show all 59 references
  1. [9]

    Bot-sort: Robust associations multi- pedestrian tracking

    Nir Aharon, Roy Orfaig, and Ben-Zion Bobrovsky. Bot-sort: Robust associations multi- pedestrian tracking. arXiv preprint arXiv:2206.14651, 2022

    2022 arXiv

  2. [10]

    Hard to track objects with irregular motions and similar appearances? make it easier by buffering the matching space

    Fan Yang, Shigeyuki Odashima, Shoichi Masui, and Shan Jiang. Hard to track objects with irregular motions and similar appearances? make it easier by buffering the matching space. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 4799–4808, 2023

    2023

  3. [11]

    Denoising-mot: Towards multiple object tracking with severe occlusions

    Teng Fu, Xiaocong Wang, Haiyang Yu, Ke Niu, Bin Li, and Xiangyang Xue. Denoising-mot: Towards multiple object tracking with severe occlusions. In Proceedings of the 31st ACM International Conference on Multimedia, pages 2734–2743, 2023

    2023

  4. [12]

    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

  5. [13]

    Yolox: Exceeding yolo series in

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. Yolox: Exceeding yolo series in

  6. [14]

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

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28, 2015

    2015

  7. [15]

    A new approach to linear filtering and prediction problems

    Rudolph Emil Kalman. A new approach to linear filtering and prediction problems. 1960

    1960

  8. [16]

    Fastreid: a pytorch toolbox for real-world person re-identification

    Lingxiao He, Xingyu Liao, Wu Liu, Xinchen Liu, Peng Cheng, and Tao Mei. Fastreid: a pytorch toolbox for real-world person re-identification. arXiv preprint arXiv:2006.02631, 1(7):6, 2020

    2006 arXiv

  9. [17]

    Mot16: A benchmark for multi-object tracking

    Anton Milan, Laura Leal-Taixé, Ian Reid, Stefan Roth, and Konrad Schindler. Mot16: A benchmark for multi-object tracking. arXiv preprint arXiv:1603.00831, 2016

    2016 arXiv

  10. [18]

    Mot20: A benchmark for multi object tracking in crowded scenes

    Patrick Dendorfer, Hamid Rezatofighi, Anton Milan, Javen Shi, Daniel Cremers, Ian Reid, Stefan Roth, Konrad Schindler, and Laura Leal-Taixé. Mot20: A benchmark for multi object tracking in crowded scenes. arXiv preprint arXiv:2003.09003, 2020

    2003 arXiv

  11. [19]

    Crowdhuman: A benchmark for detecting human in a crowd

    Shuai Shao, Zijian Zhao, Boxun Li, Tete Xiao, Gang Yu, Xiangyu Zhang, and Jian Sun. Crowdhuman: A benchmark for detecting human in a crowd. arXiv preprint arXiv:1805.00123, 2018

    2018 arXiv

  12. [20]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In 2012 IEEE conference on computer vision and pattern recognition, pages 3354–3361. IEEE, 2012

    2012

  13. [21]

    Dance- track: Multi-object tracking in uniform appearance and diverse motion

    Peize Sun, Jinkun Cao, Yi Jiang, Zehuan Yuan, Song Bai, Kris Kitani, and Ping Luo. Dance- track: Multi-object tracking in uniform appearance and diverse motion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20993–21002, 2022

    2022

  14. [22]

    Sportsmot: A large multi-object tracking dataset in multiple sports scenes

    Yutao Cui, Chenkai Zeng, Xiaoyu Zhao, Yichun Yang, Gangshan Wu, and Limin Wang. Sportsmot: A large multi-object tracking dataset in multiple sports scenes. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9921–9931, 2023

    2023

  15. [23]

    Ovtrack: Open-vocabulary multiple object tracking

    Siyuan Li, Tobias Fischer, Lei Ke, Henghui Ding, Martin Danelljan, and Fisher Yu. Ovtrack: Open-vocabulary multiple object tracking. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5567–5577, 2023

    2023

  16. [24]

    Matching anything by segmenting anything

    Siyuan Li, Lei Ke, Martin Danelljan, Luigi Piccinelli, Mattia Segu, Luc Van Gool, and Fisher Yu. Matching anything by segmenting anything. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18963–18973, 2024

    2024

  17. [25]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017

    2017

  18. [26]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  19. [27]

    Deformable detr: Deformable transformers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020

    2010 arXiv

  20. [28]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part I 16, pages 2...

    2020

  21. [29]

    Object-centric multiple object tracking

    Zixu Zhao, Jiaze Wang, Max Horn, Yizhuo Ding, Tong He, Zechen Bai, Dominik Zietlow, Carl-Johann Simon-Gabriel, Bing Shuai, Zhuowen Tu, et al. Object-centric multiple object tracking. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16601–16611,...

    2023

  22. [30]

    Diffusiontrack: Diffusion model for multi-object tracking

    Run Luo, Zikai Song, Lintao Ma, Jinlin Wei, Wei Yang, and Min Yang. Diffusiontrack: Diffusion model for multi-object tracking. arXiv preprint arXiv:2308.09905, 2023

    2023 arXiv

  23. [31]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. arXiv preprint arXiv:2304.02643, 2023

    2023 arXiv

  24. [32]

    Decoupling features in hierarchical propagation for video object segmentation

    Zongxin Yang and Yi Yang. Decoupling features in hierarchical propagation for video object segmentation. Advances in Neural Information Processing Systems, 35:36324–36336, 2022

    2022

  25. [33]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv preprint arXiv:2303.05499, 2023

    2023 arXiv

  26. [34]

    Segment and track anything

    Yangming Cheng, Liulei Li, Yuanyou Xu, Xiaodi Li, Zongxin Yang, Wenguan Wang, and Yi Yang. Segment and track anything. arXiv preprint arXiv:2305.06558, 2023

    2023 arXiv

  27. [35]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. In European Conference on Computer Vision, pages 38–55. Springer, 2024

    2024

  28. [36]

    Visual prompt multi- modal tracking

    Jiawen Zhu, Simiao Lai, Xin Chen, Dong Wang, and Huchuan Lu. Visual prompt multi- modal tracking. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9516–9526, 2023

    2023

  29. [37]

    Mots: Multi-object tracking and segmentation

    Paul V oigtlaender, Michael Krause, Aljosa Osep, Jonathon Luiten, Berin Balachandar Gnana Sekar, Andreas Geiger, and Bastian Leibe. Mots: Multi-object tracking and segmentation. In Proceedings of the ieee/cvf conference on computer vision and pattern recognition, pages 7942–7951, 2019

    2019

  30. [38]

    Youtube-vos: A large-scale video object segmentation benchmark

    Ning Xu, Linjie Yang, Yuchen Fan, Dingcheng Yue, Yuchen Liang, Jianchao Yang, and Thomas Huang. Youtube-vos: A large-scale video object segmentation benchmark. arXiv preprint arXiv:1809.03327, 2018

    2018 arXiv

  31. [39]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, et al. Scalability in perception for autonomous driving: Waymo open dataset. In Proceedings of the IEEE/CVF conference on computer vi...

    2020

  32. [40]

    Bdd100k: A diverse driving video database with scalable annotation tooling

    Fisher Yu, Wenqi Xian, Yingying Chen, Fangchen Liu, Mike Liao, Vashisht Madhavan, Trevor Darrell, et al. Bdd100k: A diverse driving video database with scalable annotation tooling. arXiv preprint arXiv:1805.04687, 2(5):6, 2018

    2018 arXiv

  33. [41]

    Imagenet: A large- scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large- scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

    2009

  34. [42]

    Tao: A large-scale benchmark for tracking any object

    Achal Dave, Tarasha Khurana, Pavel Tokmakov, Cordelia Schmid, and Deva Ramanan. Tao: A large-scale benchmark for tracking any object. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16 , pages 436–454. Springer, 2020

    2020

  35. [43]

    Beyond mot: Semantic multi-object tracking

    Yunhao Li, Qin Li, Hao Wang, Xue Ma, Jiali Yao, Shaohua Dong, Heng Fan, and Libo Zhang. Beyond mot: Semantic multi-object tracking. In European Conference on Computer Vision, pages 276–293. Springer, 2025

    2025

  36. [44]

    Referring multi-object tracking

    Dongming Wu, Wencheng Han, Tiancai Wang, Xingping Dong, Xiangyu Zhang, and Jianbing Shen. Referring multi-object tracking. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14633–14642, 2023

    2023

  37. [45]

    Bootstrapping referring multi-object tracking

    Yani Zhang, Dongming Wu, Wencheng Han, and Xingping Dong. Bootstrapping referring multi-object tracking. arXiv preprint arXiv:2406.05039, 2024. 12

    2024

  38. [46]

    Cross-view referring multi-object tracking.arXiv preprint arXiv:2412.17807, 2024

    Sijia Chen, En Yu, and Wenbing Tao. Cross-view referring multi-object tracking.arXiv preprint arXiv:2412.17807, 2024

    2024 arXiv

  39. [47]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv...

    2010 arXiv

  40. [48]

    Evaluating multiple object tracking performance: the clear mot metrics

    Keni Bernardin and Rainer Stiefelhagen. Evaluating multiple object tracking performance: the clear mot metrics. EURASIP Journal on Image and Video Processing, 2008:1–10, 2008

    2008

  41. [49]

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

    Jonathon Luiten, Aljosa Osep, Patrick Dendorfer, Philip Torr, Andreas Geiger, Laura Leal- Taixé, and Bastian Leibe. Hota: A higher order metric for evaluating multi-object tracking. International journal of computer vision, 129:548–578, 2021

    2021

  42. [50]

    Tracking objects as points

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

    2020

  43. [51]

    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

  44. [52]

    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

  45. [53]

    Diffusion models: A comprehensive survey of methods and applications

    Ling Yang, Zhilong Zhang, Yang Song, Shenda Hong, Runsheng Xu, Yue Zhao, Wentao Zhang, Bin Cui, and Ming-Hsuan Yang. Diffusion models: A comprehensive survey of methods and applications. ACM Computing Surveys, 56(4):1–39, 2023

    2023

  46. [54]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023

    2023 arXiv

  47. [55]

    Object-centric learning with slot attention

    Francesco Locatello, Dirk Weissenborn, Thomas Unterthiner, Aravindh Mahendran, Georg Heigold, Jakob Uszkoreit, Alexey Dosovitskiy, and Thomas Kipf. Object-centric learning with slot attention. Advances in Neural Information Processing Systems, 33:11525–11538, 2020

    2020

  48. [56]

    Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models. arXiv preprint arXiv:2301.12597, 2023

    2023 arXiv

  49. [57]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

    2021

  50. [58]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9(11), 2008. 13

    2008

  51. [2021]

    arXiv preprint arXiv:2107.08430, 2021. 10

    2021 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.