Pith. sign in

REVIEW 4 major objections 7 minor 106 references

Progressive Scaling Visual Object Tracking

T0 review · 4 major / 7 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 A progressive training recipe — small teacher transfer plus masked dual-branch alignment — lets larger trackers beat state-of-the-art methods on a new 12-dataset benchmark with no added inference cost.

desk verdict Useful scaling recipe and a new benchmark, but the evaluation splits are under-specified and the headline accuracy claims overreach. read the letter →

arxiv 2505.19990 v2 pith:RRTAZUC7 submitted 2025-05-26 cs.CV

Jack Hong , Shilin Yan , Zehao Xiao , Jiayin Cai , Xiaolong Jiang , Yao Hu , Henghui Ding This is my paper · ORCID
classification cs.CV
keywords visualobjecttrackingscalinglawmodelsizetrainingdatavolumeinputresolutionDT-Trainingknowledgedistillationbenchmarkevaluation
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 is trying to establish that visual object trackers improve more when model size, training data, and input resolution are scaled up in stages than when any one of them is scaled naively. Its central result is DT-Training, a recipe in which a smaller, already-trained model stays frozen as a teacher for a larger student while a randomly masked copy of each input image is aligned with the clean image through shared weights. On GTrack Bench, a new evaluation suite of 4,369 trajectories drawn from 12 datasets, the scaled model scores 64.8 mean AUC, beating published trackers by at least 1.4. If the claim is right, scaling laws apply to tracking too, and the gain shows up without extra inference cost because the recipe changes training, not the deployed architecture.

What carries the argument

The mechanism is DT-Training, built from two losses plus an outer loop. Small teacher transfer freezes a smaller pretrained tracker and minimizes a tracking loss plus L2 feature distance between its outputs and those of the larger student, so the student inherits the optimization path of a model that already works. Dual-branch alignment feeds the same image with and without random patches through one shared-weight student and aligns the two branches' outputs and features, increasing task difficulty and robustness. Around this, the paper wraps progressive scaling: at each stage one factor — data, model size, or resolution — is increased, and the model from the previous stage becomes the teacher for the next, so gains accumulate iteratively and the ordering of the three factors is flexible.

What would settle it

Run a clean held-out audit of GTrack Bench: fix exact train/test splits for all 12 component datasets, use only the LaSOT training split for the 40-epoch fine-tune, release the checkpoint, and independently recompute the mean AUC. The claim is falsified if the clean score drops below 63.9 (the best fully reported baseline) or if the margin over that baseline falls short of the advertised 1.4.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that ordinary naive training is the bottleneck in scaling trackers: a ViT-H backbone adds only 0.3 mean AUC over ViT-L, and extra data quickly hits diminishing returns. DT-Training reverses the usual distillation direction, using a small frozen teacher to guide a larger student, and adds a dual-branch alignment loss that forces the shared-weights network to produce consistent predictions and features from clean and randomly masked inputs. The paper reports that this recipe doubles the gain of naive training when scaling from ViT-Base at 256 resolution to ViT-Large at 384 resolution on LaSOT, 4.7% versus 2.4%, and that the progressively trained model reaches 64.8 mean AUC on GTrack Bench, ahead of the strongest compared trackers by at least 1.4 mean AUC. It also reports that the same training regime transfers: a compressed variant keeps competitive AUC, replacing a backbone improves depth, thermal, and event tracking, and applying the recipe to Deformable DETR improves object detection AP by 1.5.

Load-bearing premise

The whole superiority claim rests on the premise that the 40-epoch LaSOT fine-tuning and the enlarged training collection never touch the videos used to score GTrack Bench, because if evaluation trajectories were seen during training or fine-tuning the 64.8 mean AUC would be inflated.

Editorial extensions

If this is right

  • Practitioners using DT-Training can scale a tracker's data, model size, or resolution and obtain consistent gains over naive training on each axis separately.
  • The resulting model outperforms state-of-the-art trackers on GTrack Bench while preserving the baseline's inference speed, because nothing about the deployed architecture changes.
  • The recipe transfers beyond tracking: the same training scheme raises Deformable DETR's object detection AP by 1.5 under identical settings.
  • Strong learned representations survive compression, since the scaled model still improves on CompressTracker after being squeezed into four transformer layers.
  • Progressive scaling turns diminishing returns into an iterative process, with each larger model starting from the knowledge of the previous smaller one.

Reading between the lines

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

  • Extension (ours): if the mechanism is as general as the detection experiment suggests, DT-Training should also improve other dense prediction tasks such as segmentation and depth estimation, where masked-image consistency and small-teacher guidance are natural fits.
  • Implication (ours): because each stage's teacher is the previous stage's student, returns may depend on the order of scaling; the paper states the order is flexible but does not compare different orders.
  • Testable prediction (ours): the dual-branch alignment term should matter most for robustness to partial occlusion, so a benchmark that adds controlled occlusions could separate alignment gains from plain capacity 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 / 7 minor

Summary. The paper proposes DT-Training, a progressive scaling training strategy for visual object tracking that combines small-teacher transfer and dual-branch masked alignment. The authors scale training data, model size, and input resolution progressively, introduce GTrack Bench (an aggregation of 12 existing datasets with 4,369 trajectories), and report a 64.8 mean AUC on GTrack Bench, claiming to surpass state-of-the-art methods by at least 1.4 mean AUC. Additional experiments address model compression, multimodal tracking, and transfer to object detection.

Significance. If the results held under a clean evaluation protocol, the paper would provide a useful empirical study of scaling factors in visual tracking and a simple training recipe that transfers across architectures and modalities. The paper is transparent about several training details, such as loss hyperparameters in Sec. 3.5 and the dataset composition in Appendix A.2, and the compression and multimodal experiments are positive contributions. However, the central SOTA claim currently rests on an evaluation protocol whose train/test separation is not established, on a mean aggregation rule that is not stated, and on a comparison set that omits strong partial results. These issues are load-bearing for the headline claims.

major comments (4)
  1. [Secs. 4.1, 4.3, A.2; Table 1] The GTrack Bench evaluation protocol is not clean. GTrack Bench includes MOSE and OVIS (531 and 859 trajectories, respectively, in Table 1), and Appendix A.2 lists MOSE and OVIS among the training datasets, without stating which trajectories are used for training versus evaluation. Similarly, Sec. 4.3 says the scaled model is 'finetune[d] on LaSOT for 40 epochs' without specifying whether this uses the LaSOT training split or the full dataset, and LaSOT is also a GTrack Bench component. If evaluation trajectories were seen during training or fine-tuning, the reported 64.8 mean AUC is inflated. The paper must define exact train/test splits for every constituent dataset and release the split metadata before the headline claim can be verified.
  2. [Tables 3 and 4; Sec. 4.2] The mean AUC column cannot be reproduced from the per-benchmark values shown. For Ours-L-384-M, the arithmetic mean of the 12 reported values is 61.6, not 64.8; the reported 64.8 corresponds to a trajectory-weighted average using the trajectory counts in Table 1, but the weighting is never stated. Moreover, under the same trajectory weighting, ARTrack-L384 has a mean of 63.9, not 63.4, so the abstract's claim of 'at least 1.4 mean AUC' over state-of-the-art methods is not supported by the table's own values: the margin over ARTrack-L384 is about 0.8 mean AUC. The aggregation formula and the comparison set must be specified, and the claimed margins must be recomputed.
  3. [Abstract and Sec. 1 vs. Table 2] The claimed 4.7% improvement on LaSOT when scaling from ViT-Base to ViT-Large at 384 resolution is confounded. The comparison of Ours-L-384-M (73.1 AUC) against Baseline-B-256-N (68.4 AUC) simultaneously changes model size, input resolution, and training data, and it includes the progressive scaling recipe. The single-factor rows in Table 2 show DT-Training gains of only 0.5 to 0.8 AUC over the corresponding naive baseline, so the 4.7-point gap cannot be attributed to DT-Training 'doubling the gain' of naive scaling. The scaling comparison should isolate the effect of DT-Training at each factor.
  4. [Table 4] The state-of-the-art comparison is incomplete in a way that directly affects the headline. ARTrackV2-L384 and LoRAT-L-378 are listed with per-benchmark values but no mean, despite LoRAT-L-378 numerically outperforming Ours-L-384-M on every common benchmark it reports (75.1 vs. 73.1 on LaSOT, 56.6 vs. 53.0 on LaSOText, 85.6 vs. 84.7 on TrackingNet, 62.3 vs. 66.3 on TNL2K? correction: TNL2K 62.3 vs. 66.3 is a loss for LoRAT; the first two are wins). The paper should either report full GTrack Bench means for these methods using the same aggregation rule, or explicitly state why their partial results are excluded. As written, the claim that the proposed model 'surpasses existing methods by at least 1.4 mean AUC' is not established.
minor comments (7)
  1. [Sec. 4.3] The text says the model has 'a performance gain of at least 2% in the average AUC over ten benchmarks,' but GTrack Bench has 12 constituent datasets; this should read 'twelve benchmarks.'
  2. [Tables 2, 3, and 4] The same baseline model is reported inconsistently: Baseline-L-256-N has LaSOT AUC 70.0 in Tables 2 and 3 but 69.9 in Table 4, and LaSOText AUC 46.6 in Tables 2 and 3 but 47.1 in Table 4. The authors should ensure identical models receive identical numbers across tables.
  3. [Abstract and Sec. 1] The phrases '4.7% improvement' and '2.4%' should be expressed as AUC point differences (4.7 and 2.4 points) to avoid confusion with relative percentage gains.
  4. [Sec. 3.3] The sentence 'we first utilize use groundtruth supervision' should read 'we first use ground-truth supervision.'
  5. [Sec. 4.2] The text says 'Please see Supplementary Materials for more details about our GTrack Bench,' but no supplementary material is provided in the manuscript; the benchmark construction, per-dataset evaluation protocols, and train/test splits should be included in the paper.
  6. [Figure 3 caption] The caption contains a typo: 'mask ration' should be 'mask ratio.'
  7. [Sec. 5.1] The phrase 'origin Deformable DETR' should be 'original Deformable DETR.'

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: DT-Training is an empirical recipe whose reported gains are not defined from the benchmark scores; evaluation-split ambiguity is a protocol risk, not a circularity.

full rationale

The paper's derivation chain is empirical rather than formal: Eq. (5) defines Ltotal as a supervised tracking loss plus teacher-transfer and masked-branch alignment terms, and the reported AUC numbers (Tables 2-4) are outcomes of training, not quantities solved for from those equations. No fitted parameter is renamed as a prediction, and no uniqueness theorem or ansatz is imported via self-citation. Self-references (CompressTracker [38], OneTracker [39], MOSE [25]) serve as evaluation frameworks or benchmark components, not as evidence that forces the central scaling claim. The one legitimate concern is evaluation protocol: GTrack Bench (Table 1) includes LaSOT, TNL2K, MOSE, and OVIS, while Appendix A.2/Table 9 list the same datasets in the training mixture, and Sec. 4.3 says only 'we finetune the scaled model on LaSOT for 40 epochs' without stating the split. If full datasets including evaluation trajectories were used, the headline AUC would be inflated; however, standard tracking practice uses official train splits, and the manuscript does not exhibit the specific reduction that would make the reported score equal to a training objective by construction. This is an experimental-validity/reproducibility risk, not a circularity in the claimed derivation, so the circularity score is low.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central claims rest on manually chosen hyperparameters (lambda weights, mask schedule, training budget, LaSOT fine-tune length) and on assumptions about benchmark comparability, dataset conversion, and the representativeness of OSTrack. No new physical or conceptual entities are introduced.

free parameters (5)
  • lambda_transfer (teacher-transfer weight) = 0.5 for epochs 1-270, 0 for epochs 271-300
    Weight on the teacher-transfer loss in Eq. (5); chosen by hand, sensitivity shown in Fig. 3.
  • lambda_align (dual-branch alignment weight) = 0.1
    Weight on the masked/clean alignment loss in Eq. (5); set manually, with an ablation in Fig. 3.
  • mask_ratio_schedule = increased from 0.05 to 0.4
    Masking probability for the dual-branch input; the schedule is a manually chosen curriculum.
  • finetune_epochs_on_LaSOT = 40
    The scaled model is fine-tuned on LaSOT before GTrack Bench evaluation (Sec. 4.3); the train/test split is not stated.
  • training_epochs_and_batch_size = 300 epochs, batch size 256
    Training budget used for all scaling stages; chosen by hand.
assumptions (3)
  • domain assumption Averaging AUC across the 12 GTrack Bench datasets produces a comparable single performance metric.
    The constituent benchmarks differ in length, resolution, and protocol; the paper averages them without re-normalization.
  • domain assumption Converted trajectories from MOT, VOS, VIS, and open-world tracking are valid single-object tracking training examples.
    Sec. 4.1/A.2 describes the conversion at a high level, with no precise protocol, so label validity is assumed.
  • domain assumption OSTrack is a representative tracker, so scaling conclusions transfer to other architectures.
    All pioneer experiments and the scaled models use OSTrack; transfer to other trackers is asserted, not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Progressive Scaling Visual Object Tracking." pith.science (2026). https://pith.science/paper/RRTAZUC7

@misc{pith2026250519990,
  author       = {Pith},
  title        = {Pith review of: Progressive Scaling Visual Object Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RRTAZUC7}},
  note         = {Machine review of arXiv:2505.19990}
}
read the original abstract

In this work, we propose a progressive scaling training strategy for visual object tracking, systematically analyzing the influence of training data volume, model size, and input resolution on tracking performance. Our empirical study reveals that while scaling each factor leads to significant improvements in tracking accuracy, naive training suffers from suboptimal optimization and limited iterative refinement. To address this issue, we introduce DT-Training, a progressive scaling framework that integrates small teacher transfer and dual-branch alignment to maximize model potential. The resulting scaled tracker consistently outperforms state-of-the-art methods across multiple benchmarks, demonstrating strong generalization and transferability of the proposed method. Furthermore, we validate the broader applicability of our approach to additional tasks, underscoring its versatility beyond tracking.

Figures

Figures reproduced from arXiv: 2505.19990 by the authors.

Figure 1
Figure 1. Pioneer Experiments. We analyze the impact of three key factors in visual object tracking: (a) model size, (b) training data volume, and (c) input resolution. capabilities, outperforming current counterparts on GTrack Bench. Our model achieves 64.8 mean AUC, exceeding state-of-the-art methods by at least 1.4 mean AUC. Fur￾thermore, it exhibits strong transferability, maintaining high performance even after compressi… view at source ↗
Figure 2
Figure 2. Overview of our progressive scaling approach, DT-Training. Our DT-Training includes small teacher transfer and dual-branch alignment. We provide an illustrative example of continuous iterative expansion to show a gradual increase in training data, model size, and image resolution. The order of expanding the three key factors is flexible and can be adjusted as needed. larger models and fail to leverage the potential … view at source ↗
Figure 3
Figure 3. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

106 extracted references · 51 canonical work pages

  1. [1]

    Tao: A large-scale bench- mark for tracking any object.European Conference on Computer Vision, pages 436–454, 2020

    Dave Achal, Khurana Tarasha, Tokmakov Pavel, Schmid Cordelia, and Ramanan Deva. Tao: A large-scale bench- mark for tracking any object.European Conference on Computer Vision, pages 436–454, 2020. 14

    2020

  2. [2]

    Revisiting neural scaling laws in language and vi- sion.Advances in Neural Information Processing Systems, 35:22300–22312, 2022

    Ibrahim M Alabdulmohsin, Behnam Neyshabur, and Xiao- hua Zhai. Revisiting neural scaling laws in language and vi- sion.Advances in Neural Information Processing Systems, 35:22300–22312, 2022. 2

    2022

  3. [3]

    Getting vit in shape: Scal- ing laws for compute-optimal model design.Advances in Neural Information Processing Systems, 36, 2024

    Ibrahim M Alabdulmohsin, Xiaohua Zhai, Alexander Kolesnikov, and Lucas Beyer. Getting vit in shape: Scal- ing laws for compute-optimal model design.Advances in Neural Information Processing Systems, 36, 2024. 2

    2024

  4. [4]

    ARTrackV2: Prompting Autoregressive Tracker Where to Look and How to Describe

    Yifan Bai, Zeyang Zhao, Yihong Gong, and Xing Wei. Ar- trackv2: Prompting autoregressive tracker where to look and how to describe.arXiv preprint arXiv:2312.17133,

    work page Pith review arXiv

  5. [5]

    Ar- trackv2: Prompting autoregressive tracker where to look and how to describe

    Yifan Bai, Zeyang Zhao, Yihong Gong, and Xing Wei. Ar- trackv2: Prompting autoregressive tracker where to look and how to describe. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 19048–19057, 2024. 6, 7

    2024

  6. [6]

    Fully-convolutional siamese networks for object tracking

    Luca Bertinetto, Jack Valmadre, Joao F Henriques, Andrea Vedaldi, and Philip HS Torr. Fully-convolutional siamese networks for object tracking. InComputer Vision–ECCV 2016 Workshops: Amsterdam, The Netherlands, October 8-10 and 15-16, 2016, Proceedings, Part II 14, pages 850–

    2016

  7. [7]

    Learning discriminative model prediction for tracking

    Goutam Bhat, Martin Danelljan, Luc Van Gool, and Radu Timofte. Learning discriminative model prediction for tracking. InProceedings of the IEEE/CVF international conference on computer vision, pages 6182–6191, 2019

    2019

  8. [8]

    Visual object tracking using adaptive corre- lation filters

    David S Bolme, J Ross Beveridge, Bruce A Draper, and Yui Man Lui. Visual object tracking using adaptive corre- lation filters. In2010 IEEE computer society conference on computer vision and pattern recognition, pages 2544–2550. IEEE, 2010. 2

    2010

Show all 106 references
  1. [9]

    Language models are few-shot learners

    Tom B Brown. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020. 2

    2005 arXiv

  2. [10]

    Smpler-x: Scaling up expressive human pose and shape estimation.Advances in Neural Information Processing Systems, 36, 2024

    Zhongang Cai, Wanqi Yin, Ailing Zeng, Chen Wei, Qingping Sun, Wang Yanjun, Hui En Pang, Haiyi Mei, Mingyuan Zhang, Lei Zhang, et al. Smpler-x: Scaling up expressive human pose and shape estimation.Advances in Neural Information Processing Systems, 36, 2024. 2

    2024

  3. [11]

    Back- bone is all your need: A simplified architecture for visual object tracking

    Boyu Chen, Peixia Li, Lei Bai, Lei Qiao, Qiuhong Shen, Bo Li, Weihao Gan, Wei Wu, and Wanli Ouyang. Back- bone is all your need: A simplified architecture for visual object tracking. InEuropean Conference on Computer Vi- sion, pages 375–392. Springer, 2022. 2

    2022

  4. [12]

    Deepdriving: Learning affordance for direct percep- tion in autonomous driving

    Chenyi Chen, Ari Seff, Alain Kornhauser, and Jianxiong Xiao. Deepdriving: Learning affordance for direct percep- tion in autonomous driving. InProceedings of the IEEE international conference on computer vision, pages 2722– 2730, 2015. 1

    2015

  5. [13]

    Transformer tracking

    Xin Chen, Bin Yan, Jiawen Zhu, Dong Wang, Xiaoyun Yang, and Huchuan Lu. Transformer tracking. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8126–8135, 2021. 2, 8

    2021

  6. [14]

    Seqtrack: Sequence to sequence learning for vi- sual object tracking

    Xin Chen, Houwen Peng, Dong Wang, Huchuan Lu, and Han Hu. Seqtrack: Sequence to sequence learning for vi- sual object tracking. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 14572–14581, 2023. 2, 6, 7

    2023

  7. [15]

    Siamese box adaptive network for visual tracking

    Zedu Chen, Bineng Zhong, Guorong Li, Shengping Zhang, and Rongrong Ji. Siamese box adaptive network for visual tracking. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6668–6677,

  8. [16]

    Reproducible scaling laws for contrastive language-image learning

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, and Jenia Jitsev. Reproducible scaling laws for contrastive language-image learning. InProceedings of the IEEE/CVF Conference on Computer Vision ...

    2023

  9. [17]

    Mixformer: End-to-end tracking with iterative mixed atten- tion

    Yutao Cui, Cheng Jiang, Limin Wang, and Gangshan Wu. Mixformer: End-to-end tracking with iterative mixed atten- tion. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 13608–13618,

  10. [18]

    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. 14

    2023

  11. [19]

    Mixformerv2: Efficient fully transformer tracking.Ad- vances in Neural Information Processing Systems, 36,

    Yutao Cui, Tianhui Song, Gangshan Wu, and Limin Wang. Mixformerv2: Efficient fully transformer tracking.Ad- vances in Neural Information Processing Systems, 36,

  12. [20]

    High-performance long- term tracking with meta-updater

    Kenan Dai, Yunhua Zhang, Dong Wang, Jianhua Li, Huchuan Lu, and Xiaoyun Yang. High-performance long- term tracking with meta-updater. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6298–6307, 2020. 8

    2020

  13. [21]

    Atom: Accurate tracking by overlap maximization

    Martin Danelljan, Goutam Bhat, Fahad Shahbaz Khan, and Michael Felsberg. Atom: Accurate tracking by overlap maximization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4660– 4669, 2019. 2

    2019

  14. [22]

    Scaling vision transformers to 22 billion pa- rameters

    Mostafa Dehghani, Josip Djolonga, Basil Mustafa, Piotr Padlewski, Jonathan Heek, Justin Gilmer, Andreas Peter Steiner, Mathilde Caron, Robert Geirhos, Ibrahim Alabdul- mohsin, et al. Scaling vision transformers to 22 billion pa- rameters. InInternational Conference on Machine ...

    2023

  15. [23]

    Mot20: A benchmark for multi object tracking in crowded scenes.arXiv preprint arXiv:2003.09003, 2020

    P Dendorfer. Mot20: A benchmark for multi object tracking in crowded scenes.arXiv preprint arXiv:2003.09003, 2020. 14

    2003 arXiv

  16. [24]

    Motchallenge: A benchmark for single- camera multiple target tracking.International Journal of Computer Vision, 129:845–881, 2021

    Patrick Dendorfer, Aljosa Osep, Anton Milan, Konrad Schindler, Daniel Cremers, Ian Reid, Stefan Roth, and Laura Leal-Taix´e. Motchallenge: A benchmark for single- camera multiple target tracking.International Journal of Computer Vision, 129:845–881, 2021. 14

    2021

  17. [25]

    Mose: A new dataset for video object segmentation in complex scenes

    Henghui Ding, Chang Liu, Shuting He, Xudong Jiang, Philip HS Torr, and Song Bai. Mose: A new dataset for video object segmentation in complex scenes. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 20224–20234, 2023. 14

    2023

  18. [26]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020. 3

    2010 arXiv

  19. [27]

    The unmanned aerial vehicle benchmark: Ob- ject detection and tracking

    Dawei Du, Yuankai Qi, Hongyang Yu, Yifan Yang, Kai- wen Duan, Guorong Li, Weigang Zhang, Qingming Huang, and Qi Tian. The unmanned aerial vehicle benchmark: Ob- ject detection and tracking. InProceedings of the Euro- pean conference on computer vision (ECCV), pages 370– 386, 2018. 14

    2018

  20. [28]

    Lasot: A high-quality benchmark for large-scale single ob- ject tracking

    Heng Fan, Liting Lin, Fan Yang, Peng Chu, Ge Deng, Sijia Yu, Hexin Bai, Yong Xu, Chunyuan Liao, and Haibin Ling. Lasot: A high-quality benchmark for large-scale single ob- ject tracking. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages...

    2019

  21. [29]

    Data determines distributional robustness in contrastive lan- guage image pre-training (clip)

    Alex Fang, Gabriel Ilharco, Mitchell Wortsman, Yuhao Wan, Vaishaal Shankar, Achal Dave, and Ludwig Schmidt. Data determines distributional robustness in contrastive lan- guage image pre-training (clip). InInternational Con- ference on Machine Learning, pages 6216–6234. PMLR,

  22. [30]

    Instructseq: Unifying vision tasks with instruction- conditioned multi-modal sequence generation.arXiv preprint arXiv:2311.18835, 2023

    Rongyao Fang, Shilin Yan, Zhaoyang Huang, Jingqiu Zhou, Hao Tian, Jifeng Dai, and Hongsheng Li. Instructseq: Unifying vision tasks with instruction- conditioned multi-modal sequence generation.arXiv preprint arXiv:2311.18835, 2023. 2

    2023 arXiv

  23. [31]

    Gener- alized relation modeling for transformer tracking

    Shenyuan Gao, Chunluan Zhou, and Jun Zhang. Gener- alized relation modeling for transformer tracking. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18686–18695, 2023. 2, 6

    2023

  24. [32]

    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. In2012 IEEE conference on computer vision and pattern recognition, pages 3354–3361. IEEE, 2012. 1

    2012

  25. [33]

    Separable self and mixed attention transformers for efficient object track- ing

    Goutam Yelluru Gopal and Maria A Amer. Separable self and mixed attention transformers for efficient object track- ing. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 6708–6717, 2024. 7

    2024

  26. [34]

    High-speed tracking with kernelized correlation filters.IEEE transactions on pattern analysis and machine intelligence, 37(3):583–596, 2014

    Jo ˜ao F Henriques, Rui Caseiro, Pedro Martins, and Jorge Batista. High-speed tracking with kernelized correlation filters.IEEE transactions on pattern analysis and machine intelligence, 37(3):583–596, 2014. 2

    2014

  27. [35]

    Deep learn- ing scaling is predictable, empirically.arXiv preprint arXiv:1712.00409, 2017

    Joel Hestness, Sharan Narang, Newsha Ardalani, Gregory Diamos, Heewoo Jun, Hassan Kianinejad, Md Mostofa Ali Patwary, Yang Yang, and Yanqi Zhou. Deep learn- ing scaling is predictable, empirically.arXiv preprint arXiv:1712.00409, 2017. 2

    2017 arXiv

  28. [36]

    Training compute-optimal large language mod- els.arXiv preprint arXiv:2203.15556, 2022

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal large language mod- els.arXiv preprint arXiv:2203.15556, 2022. 2

    2022 arXiv

  29. [37]

    Worldsense: Evaluating real-world omni- modal understanding for multimodal llms.arXiv preprint arXiv:2502.04326, 2025

    Jack Hong, Shilin Yan, Jiayin Cai, Xiaolong Jiang, Yao Hu, and Weidi Xie. Worldsense: Evaluating real-world omni- modal understanding for multimodal llms.arXiv preprint arXiv:2502.04326, 2025. 2

    2025 arXiv

  30. [38]

    General compression framework for efficient transformer object tracking.arXiv preprint arXiv:2409.17564, 2024

    Lingyi Hong, Jinglun Li, Xinyu Zhou, Shilin Yan, Pinxue Guo, Kaixun Jiang, Zhaoyu Chen, Shuyong Gao, Wei Zhang, Hong Lu, et al. General compression framework for efficient transformer object tracking.arXiv preprint arXiv:2409.17564, 2024. 2, 7, 8

    2024 arXiv

  31. [39]

    Onetracker: Unifying visual object tracking with foundation models and efficient tuning

    Lingyi Hong, Shilin Yan, Renrui Zhang, Wanyun Li, Xinyu Zhou, Pinxue Guo, Kaixun Jiang, Yiting Chen, Jinglun Li, Zhaoyu Chen, et al. Onetracker: Unifying visual object tracking with foundation models and efficient tuning. In Proceedings of the IEEE/CVF Conference on Computer V...

    2024

  32. [40]

    Global instance tracking: Locating target more like hu- mans.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(1):576–592, 2022

    Shiyu Hu, Xin Zhao, Lianghua Huang, and Kaiqi Huang. Global instance tracking: Locating target more like hu- mans.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(1):576–592, 2022. 14

    2022

  33. [41]

    Got-10k: A large high-diversity benchmark for generic object track- ing in the wild.IEEE transactions on pattern analysis and machine intelligence, 43(5):1562–1577, 2019

    Lianghua Huang, Xin Zhao, and Kaiqi Huang. Got-10k: A large high-diversity benchmark for generic object track- ing in the wild.IEEE transactions on pattern analysis and machine intelligence, 43(5):1562–1577, 2019. 5, 6, 14

    2019

  34. [42]

    Automatic foveation for video compression us- ing a neurobiological model of visual attention.IEEE trans- actions on image processing, 13(10):1304–1318, 2004

    Laurent Itti. Automatic foveation for video compression us- ing a neurobiological model of visual attention.IEEE trans- actions on image processing, 13(10):1304–1318, 2004. 1

    2004

  35. [43]

    Scaling up visual and vision-language representation learning with noisy text supervision

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. Scaling up visual and vision-language representation learning with noisy text supervision. InIn- ternational conference on machine learning, pages 4904–

  36. [44]

    Exploring lightweight hierarchical vision transformers for efficient visual tracking

    Ben Kang, Xin Chen, Dong Wang, Houwen Peng, and Huchuan Lu. Exploring lightweight hierarchical vision transformers for efficient visual tracking. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 9612–9621, 2023. 7

    2023

  37. [45]

    Scaling laws for neural language models.arXiv preprint arXiv:2001.08361,

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.arXiv preprint arXiv:2001.08361,

    2001 arXiv

  38. [46]

    Big transfer (bit): General visual representation learning

    Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Joan Puigcerver, Jessica Yung, Sylvain Gelly, and Neil Houlsby. Big transfer (bit): General visual representation learning. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Pa...

    2020

  39. [47]

    High performance visual tracking with siamese region pro- posal network

    Bo Li, Junjie Yan, Wei Wu, Zheng Zhu, and Xiaolin Hu. High performance visual tracking with siamese region pro- posal network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 8971–8980,

  40. [48]

    Siamrpn++: Evolution of siamese vi- sual tracking with very deep networks

    Bo Li, Wei Wu, Qiang Wang, Fangyi Zhang, Junliang Xing, and Junjie Yan. Siamrpn++: Evolution of siamese vi- sual tracking with very deep networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4282–4291, 2019. 1, 2

    2019

  41. [49]

    Lasher: A large-scale high-diversity benchmark for RGBT tracking.IEEE Trans- actions on Image Processing, 31:392–404, 2021

    Chenglong Li, Wanlin Xue, Yaqing Jia, Zhichen Qu, Bin Luo, Jin Tang, and Dengdi Sun. Lasher: A large-scale high-diversity benchmark for RGBT tracking.IEEE Trans- actions on Image Processing, 31:392–404, 2021. 8

    2021

  42. [50]

    Tracking meets lora: Faster training, larger model, stronger performance

    Liting Lin, Heng Fan, Zhipeng Zhang, Yaowei Wang, Yong Xu, and Haibin Ling. Tracking meets lora: Faster training, larger model, stronger performance. InEuropean Confer- ence on Computer Vision, pages 300–318. Springer, 2024. 6

    2024

  43. [51]

    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. InComputer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

    2014

  44. [52]

    Neural scaling laws on graphs

    Jingzhe Liu, Haitao Mao, Zhikai Chen, Tong Zhao, Neil Shah, and Jiliang Tang. Neural scaling laws on graphs. arXiv preprint arXiv:2402.02054, 2024. 2

    2024 arXiv

  45. [53]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 6

    2017 arXiv

  46. [54]

    Ee-mllm: A data-efficient and compute- efficient multimodal large language model.arXiv preprint arXiv:2408.11795, 2024

    Feipeng Ma, Yizhou Zhou, Zheyu Zhang, Shilin Yan, Hebei Li, Zilong He, Siying Wu, Fengyun Rao, Yueyi Zhang, and Xiaoyan Sun. Ee-mllm: A data-efficient and compute- efficient multimodal large language model.arXiv preprint arXiv:2408.11795, 2024. 2

    2024 arXiv

  47. [55]

    Be- yond sot: Tracking multiple generic objects at once

    Christoph Mayer, Martin Danelljan, Ming-Hsuan Yang, Vittorio Ferrari, Luc Van Gool, and Alina Kuznetsova. Be- yond sot: Tracking multiple generic objects at once. In Proceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision, pages 6826–6836, 2024. 14

    2024

  48. [56]

    Mot16: A benchmark for multi-object tracking.arXiv preprint arXiv:1603.00831, 2016

    Anton Milan, Laura Leal-Taix ´e, Ian Reid, Stefan Roth, and Konrad Schindler. Mot16: A benchmark for multi-object tracking.arXiv preprint arXiv:1603.00831, 2016. 14

    2016 arXiv

  49. [57]

    Scaling open-vocabulary object detection.Advances in Neural Information Processing Systems, 36, 2024

    Matthias Minderer, Alexey Gritsenko, and Neil Houlsby. Scaling open-vocabulary object detection.Advances in Neural Information Processing Systems, 36, 2024. 2

    2024

  50. [58]

    A benchmark and simulator for uav tracking

    Matthias Mueller, Neil Smith, and Bernard Ghanem. A benchmark and simulator for uav tracking. InComputer Vision–ECCV 2016: 14th European Conference, Amster- dam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14, pages 445–461. Springer, 2016. 14, 15

    2016

  51. [59]

    Trackingnet: A large-scale dataset and benchmark for object tracking in the wild

    Matthias Muller, Adel Bibi, Silvio Giancola, Salman Al- subaihi, and Bernard Ghanem. Trackingnet: A large-scale dataset and benchmark for object tracking in the wild. In Proceedings of the European conference on computer vi- sion (ECCV), pages 300–317, 2018. 5, 6, 14, 15

    2018

  52. [60]

    Learning multi- domain convolutional neural networks for visual tracking

    Hyeonseob Nam and Bohyung Han. Learning multi- domain convolutional neural networks for visual tracking. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4293–4302, 2016. 8

    2016

  53. [61]

    Avist: A benchmark for vi- sual object tracking in adverse visibility.arXiv preprint arXiv:2208.06888, 2022

    Mubashir Noman, Wafa Al Ghallabi, Daniya Najiha, Christoph Mayer, Akshay Dudhane, Martin Danelljan, Hisham Cholakkal, Salman Khan, Luc Van Gool, and Fahad Shahbaz Khan. Avist: A benchmark for vi- sual object tracking in adverse visibility.arXiv preprint arXiv:2208.06888, 2022. 14

    2022 arXiv

  54. [62]

    Combined scal- ing for zero-shot transfer learning.Neurocomputing, 555: 126658, 2023

    Hieu Pham, Zihang Dai, Golnaz Ghiasi, Kenji Kawaguchi, Hanxiao Liu, Adams Wei Yu, Jiahui Yu, Yi-Ting Chen, Minh-Thang Luong, Yonghui Wu, et al. Combined scal- ing for zero-shot transfer learning.Neurocomputing, 555: 126658, 2023. 2

    2023

  55. [63]

    Occluded video instance segmentation: A benchmark.International Journal of Computer Vision, 130 (8):2022–2039, 2022

    Jiyang Qi, Yan Gao, Yao Hu, Xinggang Wang, Xiaoyu Liu, Xiang Bai, Serge Belongie, Alan Yuille, Philip HS Torr, and Song Bai. Occluded video instance segmentation: A benchmark.International Journal of Computer Vision, 130 (8):2022–2039, 2022. 14

    2022

  56. [64]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. InInternational conference on machine learning,...

    2021

  57. [65]

    Hierarchical text-conditional image generation with clip latents.arXiv preprint arXiv:2204.06125, 1(2):3, 2022

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents.arXiv preprint arXiv:2204.06125, 1(2):3, 2022. 2

    2022 arXiv

  58. [66]

    Scaling vision with sparse mix- ture of experts.Advances in Neural Information Processing Systems, 34:8583–8595, 2021

    Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, Andr´e Susano Pinto, Daniel Keysers, and Neil Houlsby. Scaling vision with sparse mix- ture of experts.Advances in Neural Information Processing Systems, 34:8583–8595, 2021. 2

    2021

  59. [67]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2

    2022

  60. [68]

    Hoot: Heavy occlusions in object tracking benchmark

    Gozde Sahin and Laurent Itti. Hoot: Heavy occlusions in object tracking benchmark. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 4830–4839, 2023. 14

    2023

  61. [69]

    Revisiting unreasonable effectiveness of data in deep learning era

    Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhi- nav Gupta. Revisiting unreasonable effectiveness of data in deep learning era. InProceedings of the IEEE international conference on computer vision, pages 843–852, 2017. 2

    2017

  62. [70]

    Dancetrack: Multi-object track- ing in uniform appearance and diverse motion

    Peize Sun, Jinkun Cao, Yi Jiang, Zehuan Yuan, Song Bai, Kris Kitani, and Ping Luo. Dancetrack: Multi-object track- ing in uniform appearance and diverse motion. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20993–21002, 2022. 14

    2022

  63. [71]

    Scale ef- ficiently: Insights from pre-training and fine-tuning trans- formers.arXiv preprint arXiv:2109.10686, 2021

    Yi Tay, Mostafa Dehghani, Jinfeng Rao, William Fedus, Samira Abnar, Hyung Won Chung, Sharan Narang, Dani Yogatama, Ashish Vaswani, and Donald Metzler. Scale ef- ficiently: Insights from pre-training and fine-tuning trans- formers.arXiv preprint arXiv:2109.10686, 2021. 2

    2021 arXiv

  64. [72]

    Robust and ef- ficient foreground analysis for real-time video surveillance

    Ying-Li Tian, Max Lu, and Arun Hampapur. Robust and ef- ficient foreground analysis for real-time video surveillance. In2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), pages 1182–

  65. [73]

    Llama: Open and efficient foundation language mod- els.arXiv preprint arXiv:2302.13971, 2023

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth ´ee Lacroix, Bap- tiste Rozi `ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language mod- els.arXiv preprint arXiv:2302.13971, 2023. 2

    2023 arXiv

  66. [74]

    Image captioners are scalable vision learners too.Advances in Neural Infor- mation Processing Systems, 36, 2024

    Michael Tschannen, Manoj Kumar, Andreas Steiner, Xiao- hua Zhai, Neil Houlsby, and Lucas Beyer. Image captioners are scalable vision learners too.Advances in Neural Infor- mation Processing Systems, 36, 2024. 2

    2024

  67. [75]

    Siam r-cnn: Visual tracking by re-detection

    Paul V oigtlaender, Jonathon Luiten, Philip HS Torr, and Bastian Leibe. Siam r-cnn: Visual tracking by re-detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6578–6588, 2020. 8

    2020

  68. [76]

    Unidentified video objects: A benchmark for dense, open- world segmentation

    Weiyao Wang, Matt Feiszli, Heng Wang, and Du Tran. Unidentified video objects: A benchmark for dense, open- world segmentation. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision, pages 10776– 10785, 2021. 14

    2021

  69. [77]

    Vi- sevent: Reliable object tracking via collaboration of frame and event flows.arXiv preprint arXiv:2108.05015, 2021

    Xiao Wang, Jianing Li, Lin Zhu, Zhipeng Zhang, Zhe Chen, Xin Li, Yaowei Wang, Yonghong Tian, and Feng Wu. Vi- sevent: Reliable object tracking via collaboration of frame and event flows.arXiv preprint arXiv:2108.05015, 2021. 8

    2021 arXiv

  70. [78]

    Towards more flexi- ble and accurate object tracking with natural language: Al- gorithms and benchmark

    Xiao Wang, Xiujun Shu, Zhipeng Zhang, Bo Jiang, Yaowei Wang, Yonghong Tian, and Feng Wu. Towards more flexi- ble and accurate object tracking with natural language: Al- gorithms and benchmark. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition,...

    2021

  71. [79]

    Autoregressive visual tracking

    Xing Wei, Yifan Bai, Yongchao Zheng, Dahu Shi, and Yi- hong Gong. Autoregressive visual tracking. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 9697–9706, 2023. 2, 6, 14

    2023

  72. [80]

    Online ob- ject tracking: A benchmark

    Yi Wu, Jongwoo Lim, and Ming-Hsuan Yang. Online ob- ject tracking: A benchmark. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 2411–2418, 2013. 1

    2013

  73. [81]

    Anygraph: Graph founda- tion model in the wild.arXiv preprint arXiv:2408.10700,

    Lianghao Xia and Chao Huang. Anygraph: Graph founda- tion model in the wild.arXiv preprint arXiv:2408.10700,

    arXiv

  74. [82]

    Attribute-based progressive fusion network for RGBT tracking

    Yun Xiao, Mengmeng Yang, Chenglong Li, Lei Liu, and Jin Tang. Attribute-based progressive fusion network for RGBT tracking. InAAAI, 2022. 8

    2022

  75. [83]

    Dynaprompt: Dynamic test-time prompt tuning.arXiv preprint arXiv:2501.16404, 2025

    Zehao Xiao, Shilin Yan, Jack Hong, Jiayin Cai, Xiaolong Jiang, Yao Hu, Jiayi Shen, Qi Wang, and Cees GM Snoek. Dynaprompt: Dynamic test-time prompt tuning.arXiv preprint arXiv:2501.16404, 2025. 2

    2025 arXiv

  76. [84]

    On data scaling in masked image modeling

    Zhenda Xie, Zheng Zhang, Yue Cao, Yutong Lin, Yixuan Wei, Qi Dai, and Han Hu. On data scaling in masked image modeling. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10365– 10374, 2023. 2

    2023

  77. [85]

    Multi- ple human tracking based on multi-view upper-body de- tection and discriminative learning

    Junliang Xing, Haizhou Ai, and Shihong Lao. Multi- ple human tracking based on multi-view upper-body de- tection and discriminative learning. In2010 20th Inter- national Conference on Pattern Recognition, pages 1698–

  78. [86]

    Learning spatio-temporal transformer for vi- sual tracking

    Bin Yan, Houwen Peng, Jianlong Fu, Dong Wang, and Huchuan Lu. Learning spatio-temporal transformer for vi- sual tracking. InProceedings of the IEEE/CVF interna- tional conference on computer vision, pages 10448–10457,

  79. [87]

    Depthtrack: Un- veiling the power of RGBD tracking

    Song Yan, Jinyu Yang, Jani K ¨apyl¨a, Feng Zheng, Ale ˇs Leonardis, and Joni-Kristian K¨am¨ar¨ainen. Depthtrack: Un- veiling the power of RGBD tracking. InICCV, pages 10725–10733, 2021. 8

    2021

  80. [88]

    Depthtrack: Un- veiling the power of rgbd tracking

    Song Yan, Jinyu Yang, Jani K ¨apyl¨a, Feng Zheng, Ale ˇs Leonardis, and Joni-Kristian K¨am¨ar¨ainen. Depthtrack: Un- veiling the power of rgbd tracking. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 10725–10733, 2021. 8

    2021

  81. [89]

    A sanity check for ai-generated image detection.arXiv preprint arXiv:2406.19435, 2024

    Shilin Yan, Ouxiang Li, Jiayin Cai, Yanbin Hao, Xi- aolong Jiang, Yao Hu, and Weidi Xie. A sanity check for ai-generated image detection.arXiv preprint arXiv:2406.19435, 2024. 2

    2024 arXiv

  82. [90]

    Panovos: Bridging non-panoramic and panoramic views with trans- former for video segmentation

    Shilin Yan, Xiaohao Xu, Renrui Zhang, Lingyi Hong, Wen- chao Chen, Wenqiang Zhang, and Wei Zhang. Panovos: Bridging non-panoramic and panoramic views with trans- former for video segmentation. InEuropean Conference on Computer Vision, pages 346–365. Springer, 2024

    2024

  83. [91]

    Referred by multi-modality: A unified tem- poral transformer for video object segmentation

    Shilin Yan, Renrui Zhang, Ziyu Guo, Wenchao Chen, Wei Zhang, Hongyang Li, Yu Qiao, Hao Dong, Zhongjiang He, and Peng Gao. Referred by multi-modality: A unified tem- poral transformer for video object segmentation. InPro- ceedings of the AAAI Conference on Artificial Intelligen...

    2024

  84. [92]

    Crosslmm: Decoupling long video sequences from lmms via dual cross-attention mechanisms.arXiv preprint arXiv:2505.17020, 2025

    Shilin Yan, Jiaming Han, Joey Tsai, Hongwei Xue, Rongyao Fang, Lingyi Hong, Ziyu Guo, and Ray Zhang. Crosslmm: Decoupling long video sequences from lmms via dual cross-attention mechanisms.arXiv preprint arXiv:2505.17020, 2025. 2

    2025

  85. [93]

    Prompting for multi-modal tracking

    Jinyu Yang, Zhe Li, Feng Zheng, Ales Leonardis, and Jingkuan Song. Prompting for multi-modal tracking. In ACMMM, pages 3492–3500, 2022. 8

    2022

  86. [94]

    Joint feature learning and relation modeling for tracking: A one-stream framework

    Botao Ye, Hong Chang, Bingpeng Ma, Shiguang Shan, and Xilin Chen. Joint feature learning and relation modeling for tracking: A one-stream framework. InEuropean con- ference on computer vision, pages 341–357. Springer, 2022. 1, 2, 3, 5, 6, 8, 14

    2022

  87. [95]

    Coca: Con- trastive captioners are image-text foundation models.arXiv preprint arXiv:2205.01917, 2022

    Jiahui Yu, Zirui Wang, Vijay Vasudevan, Legg Yeung, Mojtaba Seyedhosseini, and Yonghui Wu. Coca: Con- trastive captioners are image-text foundation models.arXiv preprint arXiv:2205.01917, 2022. 2

    2022 arXiv

  88. [96]

    Scaling vision transformers

    Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, and Lucas Beyer. Scaling vision transformers. InProceedings of the IEEE/CVF conference on computer vision and pat- tern recognition, pages 12104–12113, 2022. 2

    2022

  89. [97]

    Instance- level segmentation for autonomous driving with deep densely connected mrfs

    Ziyu Zhang, Sanja Fidler, and Raquel Urtasun. Instance- level segmentation for autonomous driving with deep densely connected mrfs. InProceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 669–677, 2016. 1

    2016

  90. [98]

    Ocean: Object-aware anchor-free tracking

    Zhipeng Zhang, Houwen Peng, Jianlong Fu, Bing Li, and Weiming Hu. Ocean: Object-aware anchor-free tracking. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXI 16, pages 771–787. Springer, 2020. 2

    2020

  91. [99]

    Reading rele- vant feature from global representation memory for visual object tracking.Advances in Neural Information Process- ing Systems, 36:10814–10827, 2023

    Xinyu Zhou, Pinxue Guo, Lingyi Hong, Jinglun Li, Wei Zhang, Weifeng Ge, and Wenqiang Zhang. Reading rele- vant feature from global representation memory for visual object tracking.Advances in Neural Information Process- ing Systems, 36:10814–10827, 2023. 2

    2023

  92. [100]

    Detrack: In-model latent denoising learning for visual object track- ing.arXiv preprint arXiv:2501.02467, 2025

    Xinyu Zhou, Jinglun Li, Lingyi Hong, Kaixun Jiang, Pinxue Guo, Weifeng Ge, and Wenqiang Zhang. Detrack: In-model latent denoising learning for visual object track- ing.arXiv preprint arXiv:2501.02467, 2025. 2

    2025 arXiv

  93. [101]

    Visual prompt multi-modal tracking.arXiv preprint arXiv:2303.10826, 2023

    Jiawen Zhu, Simiao Lai, Xin Chen, Dong Wang, and Huchuan Lu. Visual prompt multi-modal tracking.arXiv preprint arXiv:2303.10826, 2023. 8

    2023 arXiv

  94. [102]

    Deformable detr: Deformable trans- formers for end-to-end object detection.arXiv preprint arXiv:2010.04159, 2020

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable trans- formers for end-to-end object detection.arXiv preprint arXiv:2010.04159, 2020. 2, 8

    2010 arXiv

  95. [103]

    Rgbd1k: A large-scale dataset and benchmark for rgb- d object tracking.arXiv preprint arXiv:2208.09787, 2022

    Xue-Feng Zhu, Tianyang Xu, Zhangyong Tang, Zucheng Wu, Haodong Liu, Xiao Yang, Xiao-Jun Wu, and Josef Kit- tler. Rgbd1k: A large-scale dataset and benchmark for rgb- d object tracking.arXiv preprint arXiv:2208.09787, 2022. 8

    2022 arXiv

  96. [104]

    Tiny object tracking: A large-scale dataset and a baseline.IEEE transactions on neural networks and learning systems, 2023

    Yabin Zhu, Chenglong Li, Yao Liu, Xiao Wang, Jin Tang, Bin Luo, and Zhixiang Huang. Tiny object tracking: A large-scale dataset and a baseline.IEEE transactions on neural networks and learning systems, 2023. 14 A. Appendix A.1. GTrack Bench Existing tracking models [4, 17, 19,...

    2023

  97. [106]

    Sum Trajectories 280 150 511 600 123 1664 Videos 280 150 511 600 123 1664 Mean Frames 2512 2395 441 697 1247 - Table 10.Statics of current benchmarks.Trajectories in current popular benchmarks are limited

  98. [865]

    Springer, 2016. 1, 2

    2016

Pith tools

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