Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

How Texts Help? A Fine-grained Evaluation to Reveal the Role of Language in Vision-Language Tracking

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read VLTVerse claims the best text for a vision-language tracker depends on both the challenge and the tracker, and identifies correlation coefficient, delta ratio, and fast motion as the hardest challenges.

desk verdict Useful new evaluation grid for VLT, but the per-challenge text-type rankings conflate semantic content with text length and refresh schedule, so the central 'role of language' claim needs deconfounding. read the letter →

arxiv 2411.15600 v1 pith:7ELB4PGI submitted 2024-11-23 cs.CV cs.AI

Xuchen Li , Shiyu Hu , Xiaokun Feng , Dailing Zhang , Meiqi Wu , Jing Zhang , Kaiqi Huang This is my paper · ORCID
classification cs.CVcs.AI
keywords vision-languagetrackingfine-grainedevaluationchallengefactorstextualinformationmulti-granularitysemanticsingleobjectbenchmarkanalysis
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 asks whether and how language actually helps vision-language trackers, and it argues that the answer cannot be seen with the usual single-annotation benchmarks. It introduces VLTVerse, an evaluation grid that crosses 10 challenge factors with 6 types of text, creating 60 conditions for measuring trackers. On three state-of-the-art trackers, the paper finds that correlation coefficient, delta ratio, and fast motion are the hardest challenges, and that the text type producing the best results shifts from tracker to tracker and from challenge to challenge. The intended consequence is that VLT evaluation and tracker design should treat text type and challenge factor as interacting variables, not as one fixed annotation.

What carries the argument

The evaluation space is built by crossing 10 sequence-level challenge factors (abnormal ratio, abnormal scale, blur, abnormal illumination, delta illumination, delta scale, delta blur, delta ratio, fast motion, correlation coefficient) with 6 text conditions (attribute words, initial concise, dense concise, initial detailed, dense detailed, and a blank control), yielding 60 subspaces. The blank control supports a decoupled comparison that isolates the effect of introducing text, and the combination of average performance and coefficient of variation across text types reveals where language matters most.

What would settle it

Compare trackers on the same sequences using text pairs matched for length and update schedule but differing in semantic content; if performance does not vary, the claim that text semantics drive the observed differences is falsified.

Watch

Extended reading notes

Core claim

VLTVerse is presented as the first fine-grained VLT evaluation framework that combines sequence-level challenge labels with multi-granularity text. The central empirical claim is that language has a conditional rather than fixed effect: under dynamic challenges such as correlation coefficient, delta ratio, and fast motion, tracker performance varies substantially with the text prompt, and no single text type is best. The paper further claims that specific trackers have identifiable text preferences—JointNLT performs best with short initial concise text because it truncates long inputs, UVLTrack benefits from dense concise updates, and MMTrack shows no consistent pattern—so the current practice of reporting one result with one annotation obscures the role of language.

Load-bearing premise

The analysis assumes performance differences across text conditions are due to semantic content, but the conditions also differ in text length and in whether they are refreshed every 100 frames.

Editorial extensions

If this is right

  • If the framework's findings hold, VLT benchmark reports should break results down by challenge factor and text type, since a single average can hide which text helps or hurts.
  • The three hardest factors—correlation coefficient, delta ratio, and fast motion—should become priority targets for training data and text-guided designs.
  • Text length and update schedule are first-order design variables: trackers that truncate long text need length-aware handling, while dense updates help only some architectures.
  • Measuring coefficient of variation across text types gives a per-tracker sensitivity metric that exposes over-reliance on memorized text.

Reading between the lines

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

  • The reported text-type effects could be partly driven by token length and refresh timing rather than semantic content, since attribute words average 4 words while dense detailed descriptions average hundreds; a length-matched control would separate the two.
  • If text preference is architecture-dependent, then the same framework could be used to select the optimal text type at run time, e.g., by predicting the active challenge factor and switching prompts accordingly.
  • The fixed 100-frame update schedule for dense text could be replaced by an adaptive schedule keyed to detected visual degradation, which might outperform all six fixed text types.
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 / 5 minor

Summary. The paper introduces VLTVerse, a fine-grained evaluation framework for vision-language tracking (VLT) that combines 10 sequence-level challenge factors (from SOTVerse) with 6 types of textual information (Attribute Words, Initial/Dense Concise/Detailed, and Blank), yielding 60 subspaces. The authors evaluate three SOTA VLT trackers (MMTrack, JointNLT, UVLTrack) on four datasets (OTB99 Lang, TNL2K, LaSOT, MGIT) and report per-challenge-factor and per-text-type SUC results. The central conclusions are: (1) dynamic challenges (correlation coefficient, delta ratio, fast motion) are the hardest for VLT trackers; (2) different text types affect performance differently, with the best text type varying by tracker and challenge factor; and (3) text introduction can help or hurt depending on the tracker. The paper also releases a toolkit and results.

Significance. If the findings were fully substantiated, VLTVerse would provide a valuable new perspective on VLT evaluation by jointly analyzing challenge factors and semantic input types, potentially guiding tracker design and text-annotation strategies. The paper makes a serious empirical contribution: it reuses four benchmarks, adds attribute words for MGIT, and conducts a broad evaluation across 60 subspaces with three trackers. The strongest assets are the scale of the evaluation, the transparency of the framework, and the explicit consideration of text granularity and update schedules. However, the headline claims about the role of language—especially the per-challenge 'best text type' rankings—are currently undermined by a confounded experimental design and by the absence of statistical rigor, which limits the paper's significance in its present form.

major comments (4)
  1. [Section 4.3, Table 2, Section 7.1] The six text conditions differ simultaneously in semantic content, token length (Attribute Words average 4 words, while Dense Detailed averages 175–786 words), and update schedule (only Dense conditions refresh every 100 frames). Consequently, Figure 5's per-challenge 'best text type' rankings cannot be attributed solely to semantics. The paper itself concedes a non-semantic mechanism when it explains JointNLT's preference for Initial Concise as due to 'truncation of long texts' (Sec. 7.1). To establish the role of language, the authors should add conditions that independently vary text length and update cadence, or at least perform a covariate analysis that disentangles these factors from content.
  2. [Table 1] The claimed 60-subspace evaluation is sparsely populated. Several cells contain zero test sequences (e.g., Ec8 LaSOT, Ec8 MGIT, Ec10 MGIT, Ec7 MGIT, Ec3 OTB99 test), and many others have fewer than ten test sequences (e.g., Ec5 MGIT test=4, Ec10 OTB99 test=5). The fine-grained conclusions in Figures 4–5 rest on these small and sometimes empty cells, making per-subspace SUC values and rankings unreliable. The authors should report the exact sequence counts for every subspace and restrict strong rank claims to subspaces with adequate sample sizes, or provide confidence intervals that account for small n.
  3. [Section 7.1, Figures 4–5] No variance or significance testing is reported for the average SUC values, the CV values, or the 'best text type' markers. Given the small per-cell sample sizes visible in Table 1, the ranking differences in Figure 5 may reflect noise. The authors should supply standard errors, bootstrap confidence intervals, or pairwise significance tests (e.g., paired permutation tests on sequences) for the differences that underpin the central claims, particularly the ordering of the three most challenging factors and the per-challenge optimal text types.
  4. [Section 7.2, Section 4.3] The 'Blank' condition is defined as the textual input 'The tracking target' (Sec. 4.3), not as the absence of text. Therefore, comparing other text types against Blank measures the effect of a generic noun phrase, not the transition from SOT (no language) to VLT (with language). The conclusion in Section 7.2 that text introduction improves performance for MMTrack and JointNLT should be either re-framed as a comparison against a neutral textual prompt or supported by an additional true no-text control condition.
minor comments (5)
  1. [Section 6.2] There is a typo: 'JointNTL' should be 'JointNLT'.
  2. [Section 7.1] In the paragraph on UVLTrack, 'Blue Bounding-box' and 'Delta Blue' should be 'Blur Bounding-box' and 'Delta Blur'.
  3. [Section 3, Equations (2)–(3)] The notation 'Ec1i1' and the expansion in (2) are confusing; it would help to explicitly define the ordering of the 60 subtasks and clarify the relationship between Sc, Si, and Eci.
  4. [Section 4.3, Table 2] The table lists mean word counts but not the number of sequences contributing to each cell; adding that would help the reader assess the sparsity discussed in the major comments.
  5. [Section 7.1, Figure 4] The radar charts are difficult to compare across trackers because the scale is not indicated; adding axis ticks or a common scale would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is an empirical evaluation with externally checkable benchmarks; its findings are direct measurements, not derived from or equivalent to its inputs.

full rationale

VLTVerse is an empirical evaluation study rather than a derivation: the 60 subtasks are Cartesian products of 10 challenge factors and 6 text conditions, and the central claims (challenge-factor difficulty ordering, performance coefficient of variation, per-tracker optimal text types) are measured by running MMTrack, JointNLT, and UVLTrack with their original hyperparameters. The challenge-factor definitions and thresholds are inherited from SOTVerse [15], and the text annotations follow DTVLT [23]; both are published benchmarks with overlapping authors, but they are externally checkable resources and are not used to force the paper's conclusions. No fitted constants are renamed as predictions, no equation in the paper reduces to the headline result, and no uniqueness theorem is invoked to forbid alternatives. The blank-text control provides an independent baseline, and the paper's explanation of JointNLT's preference for short text via truncation is an empirical hypothesis rather than a constructed equivalence. The text-length/update-frequency confounding noted in the reviewer context is a validity concern for the semantic-content interpretation, not a circularity, because the reported rankings are observed outcomes rather than definitional consequences. Therefore the analysis is self-contained with respect to circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims rest on challenge labels defined by SOTVerse thresholds, on text annotations partially inherited from the authors' prior benchmarks (DTVLT, attribute words), and on the representativeness of three trackers. No new physical entities are introduced; the main numerical design choices are the challenge thresholds and the 100-frame refresh interval.

free parameters (2)
  • Sequence-level challenge factor thresholds = From SOTVerse [15], not disclosed in this paper
    Sequences are classified into the ten challenge subspaces by averaging per-frame attribute values and applying thresholds from SOTVerse (Sec. 4.2). These thresholds determine which sequences enter each of the 60 evaluation cells and therefore shape every conclusion.
  • Dense text update interval = 100 frames
    Dense Concise and Dense Detailed descriptions are refreshed every 100 frames (Sec. 4.3), a hand-chosen schedule that may interact with challenge duration and could affect whether dense text appears helpful.
assumptions (4)
  • domain assumption Per-frame attribute values averaged over a sequence, then thresholded, yield meaningful sequence-level challenge factors.
    Sec. 4.2 and Appendix A.3. The entire challenge-factor space depends on this aggregation; a sequence with mixed difficulties may be mislabeled.
  • domain assumption The text annotations, including attribute words from [12], concise and detailed descriptions following DTVLT [23], and newly added MGIT attribute words, are accurate and consistent across datasets.
    Sec. 4.3. The evaluation compares trackers under these text conditions, so annotation quality directly determines the findings.
  • domain assumption MMTrack, JointNLT, and UVLTrack are representative of current VLT algorithms.
    Sec. 6.1. The paper generalizes from three models with different architectures; conclusions about the role of language may not transfer.
  • domain assumption Differences between text conditions can be interpreted as effects of semantic type, not of text length or refresh timing.
    Table 2 shows large word-count differences across text types; Sec. 7.1 itself explains JointNLT's result by truncation of long text, so semantic and length effects are entangled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of How Texts Help? A Fine-grained Evaluation to Reveal the Role of Language in Vision-Language Tracking." pith.science (2026). https://pith.science/paper/7ELB4PGI

@misc{pith2026241115600,
  author       = {Pith},
  title        = {Pith review of: How Texts Help? A Fine-grained Evaluation to Reveal the Role of Language in Vision-Language Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7ELB4PGI}},
  note         = {Machine review of arXiv:2411.15600}
}
read the original abstract

Vision-language tracking (VLT) extends traditional single object tracking by incorporating textual information, providing semantic guidance to enhance tracking performance under challenging conditions like fast motion and deformations. However, current VLT trackers often underperform compared to single-modality methods on multiple benchmarks, with semantic information sometimes becoming a "distraction." To address this, we propose VLTVerse, the first fine-grained evaluation framework for VLT trackers that comprehensively considers multiple challenge factors and diverse semantic information, hoping to reveal the role of language in VLT. Our contributions include: (1) VLTVerse introduces 10 sequence-level challenge labels and 6 types of multi-granularity semantic information, creating a flexible and multi-dimensional evaluation space for VLT; (2) leveraging 60 subspaces formed by combinations of challenge factors and semantic types, we conduct systematic fine-grained evaluations of three mainstream SOTA VLT trackers, uncovering their performance bottlenecks across complex scenarios and offering a novel perspective on VLT evaluation; (3) through decoupled analysis of experimental results, we examine the impact of various semantic types on specific challenge factors in relation to different algorithms, providing essential guidance for enhancing VLT across data, evaluation, and algorithmic dimensions. The VLTVerse, toolkit, and results will be available at \url{http://metaverse.aitestunion.com}.

Figures

Figures reproduced from arXiv: 2411.15600 by the authors.

Figure 1
Figure 1. Examples of tracking results by JointNLT [ [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. VLTVerse comprises two main components: environ [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Left: Example of challenging factors, with four static challenging factors and six dynamic challenging factors. Right: Example of textual information, providing six types of information for each video sequence, including Attribute Words, Dense Concise, Dense Detailed, Initial Concise, Initial Detailed, and Blank information. based on the number of challenge factor sequences within each dataset, providing an overall … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Radar chart of the Average Value (AV) and Coefficient of Variation (CV) of tracking performance under different textual [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Optimal semantic information for tracking performance [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Radar chart of the tracking performance difference in information across various texts versus Blank under different challenge [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visualization of tracker results combined with textual [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Enhancing Vision-Language Tracking by Effectively Converting Textual Cues into Visual Cues

    cs.CV 2024-12 conditional novelty 6.0 of 10

    CTVLT converts text descriptions into spatial heatmaps via Grounding DINO and fuses them into a visual tracker, achieving reported state-of-the-art performance on MGIT, TNL2K, and LaSOT.

Reference graph

Works this paper leans on

51 extracted references · 43 canonical work pages · cited by 1 Pith paper

  1. [1]

    Human mem- ory: A proposed system and its control processes

    Richard C Atkinson and Richard M Shiffrin. Human mem- ory: A proposed system and its control processes. In Psy- chology of learning and motivation, pages 89–195. Elsevier,

  2. [2]

    Hiptrack: Visual tracking with historical prompts

    Wenrui Cai, Qingjie Liu, and Yunhong Wang. Hiptrack: Visual tracking with historical prompts. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19258–19267, 2024. 3

    work page 2024

  3. [3]

    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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5374–5383,

  4. [4]

    Lasot: A high-quality large-scale single object tracking benchmark

    Heng Fan, Hexin Bai, Liting Lin, Fan Yang, Peng Chu, Ge Deng, Sijia Yu, Harshit, Mingzhen Huang, Juehuan Liu, et al. Lasot: A high-quality large-scale single object tracking benchmark. International Journal of Computer Vision, 129: 439–461, 2021. 2

    work page 2021

  5. [5]

    The ther- mal infrared visual object tracking vot-tir2015 challenge re- sults

    Michael Felsberg, Amanda Berg, Gustav Hager, Jorgen Ahlberg, Matej Kristan, Jiri Matas, Ales Leonardis, Luka Cehovin, Gustavo Fernandez, Tomas V ojir, et al. The ther- mal infrared visual object tracking vot-tir2015 challenge re- sults. In Proceedings of the IEEE international conference on computer vision workshops, pages 76–88, 2015. 3

    work page 2015

  6. [6]

    Robust visual object tracking with natural language region proposal network

    Qi Feng, Vitaly Ablavsky, Qinxun Bai, and Stan Sclaroff. Robust visual object tracking with natural language region proposal network. arXiv preprint arXiv:1912.02048, 1(7):8,

    arXiv 1912

  7. [7]

    Real-time visual object tracking with natural lan- guage description

    Qi Feng, Vitaly Ablavsky, Qinxun Bai, Guorong Li, and Stan Sclaroff. Real-time visual object tracking with natural lan- guage description. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 700– 709, 2020

    work page 2020

  8. [8]

    Siamese natural language tracker: Tracking by natural lan- guage descriptions with siamese trackers

    Qi Feng, Vitaly Ablavsky, Qinxun Bai, and Stan Sclaroff. Siamese natural language tracker: Tracking by natural lan- guage descriptions with siamese trackers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5847–5856, 2021. 3

    work page 2021

Show all 51 references
  1. [9]

    MemVLT: Vision-language tracking with adaptive memory- based prompts

    Xiaokun Feng, Xuchen Li, Shiyu Hu, Dailing Zhang, Meiqi Wu, Jing Zhang, Xiaotang Chen, and Kaiqi Huang. MemVLT: Vision-language tracking with adaptive memory- based prompts. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024. 3

    2024

  2. [10]

    Consistencies are all you need for semi-supervised vision-language tracking

    Jiawei Ge, Jiuxin Cao, Xuelin Zhu, Xinyu Zhang, Chang Liu, Kun Wang, and Bo Liu. Consistencies are all you need for semi-supervised vision-language tracking. In Proceed- ings of the 32nd ACM International Conference on Multime- dia, pages 1895–1904, 2024. 3

    1904

  3. [11]

    Divert more attention to vision-language tracking

    Mingzhe Guo, Zhipeng Zhang, Heng Fan, and Liping Jing. Divert more attention to vision-language tracking. In Pro- ceedings of the Advances in Neural Information Processing Systems, pages 4446–4460, 2022. 3

    2022

  4. [12]

    Divert more attention to vision-language object tracking

    Mingzhe Guo, Zhipeng Zhang, Liping Jing, Haibin Ling, and Heng Fan. Divert more attention to vision-language object tracking. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2024. 5

    2024

  5. [13]

    Global instance tracking: Locating target more like humans

    Shiyu Hu, Xin Zhao, Lianghua Huang, and Kaiqi Huang. Global instance tracking: Locating target more like humans. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 45(1):576–592, 2022. 2

    2022

  6. [14]

    A multi-modal global instance tracking benchmark (mgit): Better locating target in complex spatio-temporal and causal relationship

    Shiyu Hu, Dailing Zhang, Xiaokun Feng, Xuchen Li, Xin Zhao, Kaiqi Huang, et al. A multi-modal global instance tracking benchmark (mgit): Better locating target in complex spatio-temporal and causal relationship. Advances in Neural Information Processing Systems, 36, 2024. 2, 4, 5

    2024

  7. [15]

    Sotverse: A user- defined task space of single object tracking

    Shiyu Hu, Xin Zhao, and Kaiqi Huang. Sotverse: A user- defined task space of single object tracking. International Journal of Computer Vision, 132(3):872–930, 2024. 2, 3, 4

    2024

  8. [16]

    Rtracker: Recoverable track- ing via pn tree structured memory

    Yuqing Huang, Xin Li, Zikun Zhou, Yaowei Wang, Zhenyu He, and Ming-Hsuan Yang. Rtracker: Recoverable track- ing via pn tree structured memory. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19038–19047, 2024. 3

    2024

  9. [17]

    The sixth visual object tracking vot2018 challenge results

    Matej Kristan, Ales Leonardis, Jiri Matas, Michael Fels- berg, Roman Pflugfelder, Luka ˇCehovin Zajc, Tomas V ojir, Goutam Bhat, Alan Lukezic, Abdelrahman Eldesokey, et al. The sixth visual object tracking vot2018 challenge results. In Proceedings of the European conference on...

    2018

  10. [18]

    The seventh visual object tracking vot2019 chal- lenge results

    Matej Kristan, Jiri Matas, Ales Leonardis, Michael Fels- berg, Roman Pflugfelder, Joni-Kristian Kamarainen, Luka ˇCehovin Zajc, Ondrej Drbohlav, Alan Lukezic, Amanda Berg, et al. The seventh visual object tracking vot2019 chal- lenge results. In Proceedings of the IEEE/CVF int...

    2019

  11. [19]

    The eighth visual object tracking vot2020 chal- lenge results

    Matej Kristan, Ale ˇs Leonardis, Ji ˇr´ı Matas, Michael Fels- berg, Roman Pflugfelder, Joni-Kristian K ¨am¨ar¨ainen, Martin Danelljan, Luka ˇCehovin Zajc, Alan Luke ˇziˇc, Ondrej Dr- bohlav, et al. The eighth visual object tracking vot2020 chal- lenge results. In Computer Visi...

    2020

  12. [20]

    The ninth visual object tracking vot2021 challenge results

    Matej Kristan, Ji ˇr´ı Matas, Aleˇs Leonardis, Michael Felsberg, Roman Pflugfelder, Joni-Kristian K ¨am¨ar¨ainen, Hyung Jin Chang, Martin Danelljan, Luka Cehovin, Alan Lukeˇziˇc, et al. The ninth visual object tracking vot2021 challenge results. In Proceedings of the IEEE/CVF ...

    2021

  13. [21]

    The tenth visual object tracking vot2022 challenge re- sults

    Matej Kristan, Ale ˇs Leonardis, Jiˇr´ı Matas, Michael Felsberg, Roman Pflugfelder, Joni-Kristian K ¨am¨ar¨ainen, Hyung Jin Chang, Martin Danelljan, Luka ˇCehovin Zajc, Alan Lukeˇziˇc, et al. The tenth visual object tracking vot2022 challenge re- sults. In European Conference ...

    2022

  14. [22]

    Dtllm-vlt: Diverse text generation for visual language tracking based on llm

    Xuchen Li, Xiaokun Feng, Shiyu Hu, Meiqi Wu, Dailing Zhang, Jing Zhang, and Kaiqi Huang. Dtllm-vlt: Diverse text generation for visual language tracking based on llm. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7283–7292, 2024. 2, 3

    2024

  15. [23]

    Dtvlt: A multi-modal 9 diverse text benchmark for visual language tracking based on llm

    Xuchen Li, Shiyu Hu, Xiaokun Feng, Dailing Zhang, Meiqi Wu, Jing Zhang, and Kaiqi Huang. Dtvlt: A multi-modal 9 diverse text benchmark for visual language tracking based on llm. arXiv preprint arXiv:2410.02492, 2024. 2, 3, 4, 5

    2024 arXiv

  16. [24]

    Visual language tracking with multi-modal interaction: A robust benchmark

    Xuchen Li, Shiyu Hu, Xiaokun Feng, Dailing Zhang, Meiqi Wu, Jing Zhang, and Kaiqi Huang. Visual language tracking with multi-modal interaction: A robust benchmark. arXiv preprint arXiv:2409.08887, 2024. 3

    2024 arXiv

  17. [25]

    Cross- modal target retrieval for tracking by natural language

    Yihao Li, Jun Yu, Zhongpeng Cai, and Yuwen Pan. Cross- modal target retrieval for tracking by natural language. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 4931–4940, 2022. 3

    2022

  18. [26]

    Tracking by natural language specification

    Zhenyang Li, Ran Tao, Efstratios Gavves, Cees GM Snoek, and Arnold WM Smeulders. Tracking by natural language specification. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 6495–6503,

  19. [27]

    Unifying visual and vision-language tracking via contrastive learning

    Yinchao Ma, Yuyang Tang, Wenfei Yang, Tianzhu Zhang, Jinpeng Zhang, and Mengxue Kang. Unifying visual and vision-language tracking via contrastive learning. In Pro- ceedings of the AAAI Conference on Artificial Intelligence , pages 4107–4116, 2024. 3, 6, 7, 8

    2024

  20. [28]

    Human memory

    Gabriel A Radvansky. Human memory. Routledge, 2021. 5

    2021

  21. [29]

    The visual object track- ing vot2016 challenge results

    Giorgio Roffo, Simone Melzi, et al. The visual object track- ing vot2016 challenge results. In Computer Vision–ECCV 2016 Workshops: Amsterdam, The Netherlands, October 8- 10 and 15-16, 2016, Proceedings, Part II , pages 777–823. Springer International Publishing, 2016. 3

    2016

  22. [30]

    Context-aware integration of lan- guage and visual references for natural language tracking

    Yanyan Shao, Shuting He, Qi Ye, Yuchao Feng, Wenhan Luo, and Jiming Chen. Context-aware integration of lan- guage and visual references for natural language tracking. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 19208–19217, 2024. 3

    2024

  23. [31]

    Explicit visual prompts for visual object tracking

    Liangtao Shi, Bineng Zhong, Qihua Liang, Ning Li, Sheng- ping Zhang, and Xianxian Li. Explicit visual prompts for visual object tracking. In Proceedings of the AAAI Confer- ence on Artificial Intelligence, pages 4838–4846, 2024. 3

    2024

  24. [32]

    Auditory sensory (”echoic”) memory dysfunction in schizophrenia

    Rael D Strous, Nelson Cowan, Walter Ritter, and Daniel C Javitt. Auditory sensory (”echoic”) memory dysfunction in schizophrenia. The American journal of psychiatry, 152(10): 1517–1519, 1995. 5

    1995

  25. [33]

    Chat- tracker: Enhancing visual tracking performance via chat- ting with multimodal large language model

    Yiming Sun, Fan Yu, Shaoxiang Chen, Yu Zhang, Junwei Huang, Chenhui Li, Yang Li, and Changbo Wang. Chat- tracker: Enhancing visual tracking performance via chat- ting with multimodal large language model. arXiv preprint arXiv:2411.01756, 2024. 3

    2024 arXiv

  26. [34]

    Elysium: Exploring object-level perception in videos via mllm

    Han Wang, Yongjie Ye, Yanjie Wang, Yuxiang Nie, and Can Huang. Elysium: Exploring object-level perception in videos via mllm. In European Conference on Computer Vision , pages 166–185. Springer, 2025. 2, 3

    2025

  27. [35]

    Semtrack: A large-scale dataset for semantic tracking in the wild

    Pengfei Wang, Xiaofei Hui, Jing Wu, Zile Yang, Kian Eng Ong, Xinge Zhao, Beijia Lu, Dezhao Huang, Evan Ling, Weiling Chen, et al. Semtrack: A large-scale dataset for semantic tracking in the wild. In European Conference on Computer Vision, pages 486–504. Springer, 2025. 2

    2025

  28. [36]

    Unified transformer with isomorphic branches for natural language tracking

    Rong Wang, Zongheng Tang, Qianli Zhou, Xiaoqian Liu, Tianrui Hui, Quange Tan, and Si Liu. Unified transformer with isomorphic branches for natural language tracking. IEEE Transactions on Circuits and Systems for Video Tech- nology, 2023. 3

    2023

  29. [37]

    Describe and attend to track: Learning natural language guided structural representation and visual attention for object tracking

    Xiao Wang, Chenglong Li, Rui Yang, Tianzhu Zhang, Jin Tang, and Bin Luo. Describe and attend to track: Learning natural language guided structural representation and visual attention for object tracking. arXiv preprint arXiv:1811.10014, 2018. 3

    2018 arXiv

  30. [38]

    Towards more flexible and accurate object tracking with natural language: Algo- rithms and benchmark

    Xiao Wang, Xiujun Shu, Zhipeng Zhang, Bo Jiang, Yaowei Wang, Yonghong Tian, and Feng Wu. Towards more flexible and accurate object tracking with natural language: Algo- rithms and benchmark. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition ,...

    2021

  31. [39]

    Online object tracking: A benchmark

    Yi Wu, Jongwoo Lim, and Ming-Hsuan Yang. Online object tracking: A benchmark. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 2411–2418, 2013. 2, 3

    2013

  32. [40]

    Object track- ing benchmark

    Yi Wu, Jongwoo Lim, and Ming-Hsuan Yang. Object track- ing benchmark. IEEE Transactions on Pattern Analysis and Machine Intelligence, 37(9):1834–1848, 2015. 2, 3

    2015

  33. [41]

    Autore- gressive queries for adaptive tracking with spatio-temporal transformers

    Jinxia Xie, Bineng Zhong, Zhiyi Mo, Shengping Zhang, Liangtao Shi, Shuxiang Song, and Rongrong Ji. Autore- gressive queries for adaptive tracking with spatio-temporal transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1930...

    2024

  34. [42]

    Grounding-tracking-integration

    Zhengyuan Yang, Tushar Kumar, Tianlang Chen, Jingsong Su, and Jiebo Luo. Grounding-tracking-integration. IEEE Transactions on Circuits and Systems for Video Technology, 31(9):3433–3443, 2021. 3

    2021

  35. [43]

    Webuav- 3m: A benchmark for unveiling the power of million-scale deep uav tracking

    Chunhui Zhang, Guanjie Huang, Li Liu, Shan Huang, Yinan Yang, Xiang Wan, Shiming Ge, and Dacheng Tao. Webuav- 3m: A benchmark for unveiling the power of million-scale deep uav tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(7):9186–9205, 2022. 2

    2022

  36. [44]

    Towards underwater camouflaged ob- ject tracking: An experimental evaluation of sam and sam 2

    Chunhui Zhang, Li Liu, Guanjie Huang, Hao Wen, Xi Zhou, and Yanfeng Wang. Towards underwater camouflaged ob- ject tracking: An experimental evaluation of sam and sam 2. arXiv preprint arXiv:2409.16902, 2024. 2

    2024 arXiv

  37. [45]

    Webuot-1m: Advancing deep underwa- ter object tracking with a million-scale benchmark

    Chunhui Zhang, Li Liu, Guanjie Huang, Hao Wen, Xi Zhou, and Yanfeng Wang. Webuot-1m: Advancing deep underwa- ter object tracking with a million-scale benchmark. arXiv preprint arXiv:2405.19818, 2024. 2

    2024 arXiv

  38. [46]

    One-stream stepwise decreas- ing for vision-language tracking

    Guangtong Zhang, Bineng Zhong, Qihua Liang, Zhiyi Mo, Ning Li, and Shuxiang Song. One-stream stepwise decreas- ing for vision-language tracking. IEEE Transactions on Cir- cuits and Systems for Video Technology, 2024. 3

    2024

  39. [47]

    One-stream vision-language memory network for object tracking

    Huanlong Zhang, Jingchao Wang, Jianwei Zhang, Tianzhu Zhang, and Bineng Zhong. One-stream vision-language memory network for object tracking. IEEE Transactions on Multimedia, 2023

    2023

  40. [48]

    Transformer vision-language tracking via proxy token guided cross-modal fusion

    Haojie Zhao, Xiao Wang, Dong Wang, Huchuan Lu, and Xi- ang Ruan. Transformer vision-language tracking via proxy token guided cross-modal fusion. Pattern Recognition Let- ters, 168:10–16, 2023. 3

    2023

  41. [49]

    Towards unified token learn- ing for vision-language tracking

    Yaozong Zheng, Bineng Zhong, Qihua Liang, Guorong Li, Rongrong Ji, and Xianxian Li. Towards unified token learn- ing for vision-language tracking. IEEE Transactions on Cir- cuits and Systems for Video Technology, 2023. 3, 6, 7, 8 10

    2023

  42. [50]

    Odtrack: Online dense temporal token learning for visual tracking

    Yaozong Zheng, Bineng Zhong, Qihua Liang, Zhiyi Mo, Shengping Zhang, and Xianxian Li. Odtrack: Online dense temporal token learning for visual tracking. In Proceed- ings of the AAAI Conference on Artificial Intelligence, pages 7588–7596, 2024. 3

    2024

  43. [51]

    Joint vi- sual grounding and tracking with natural language specifica- tion

    Li Zhou, Zikun Zhou, Kaige Mao, and Zhenyu He. Joint vi- sual grounding and tracking with natural language specifica- tion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 23151–23160,

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.