Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Dynamic Token Selection for Aerial-Ground Person Re-Identification

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 Dynamic token selection lets an aerial-ground person re-identification transformer beat its full-token baseline on both CARGO and AG-ReID.

desk verdict The paper's central token-selection mechanism is undefined by its own equations, so the reported gains can't be attributed to it. read the letter →

arxiv 2412.00433 v2 pith:EKXFJO3A submitted 2024-11-30 cs.CV

Yuhai Wang , Maryam Pishgar This is my paper · ORCID
classification cs.CV
keywords aerial-groundpersonre-identificationtokenselectionvisiontransformerGumbel-Softmaxtop-kview-decoupledCARGOAG-ReID
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

This paper asks whether an aerial-ground person re-identification model can match people across cameras better by throwing away most of the image and attending only to the most informative patches. It proposes the Dynamic Token Selective Transformer (DTST), which scores each image token with a lightweight attention predictor, keeps the top two tokens, and feeds only those into a view-decoupled transformer. On the CARGO dataset's aerial-to-ground protocol the paper reports 50.63% Rank-1 and 43.39% mean average precision, beating the VDT baseline by 3.13 points in Rank-1 and 1.18 points in mAP; on AG-ReID it reports Rank-1 gains of 0.57 points (aerial to ground) and 1.04 points (ground to aerial). The paper's claim is that dynamically dropping redundant tokens is what drives these gains, rather than the baseline's view-decoupling machinery.

What carries the argument

The load-bearing object is the Visual Token Selector (VTS) inserted after the final transformer block. It computes a relevance score per token, $s_i$, as a softmax over a query-key product, then picks the top-$K$ tokens; a Gumbel-Softmax reweighting is used as a continuous relaxation so gradients can flow through the discrete selection. The selector's two hyperparameters, $K=2$ and a two-head attention predictor, are the settings the ablations identify as best. The VTS sits inside the View-Decoupled Transformer's token stream, so the selected tokens inherit the meta/view token subtraction and orthogonal-loss supervision that the baseline uses.

What would settle it

Take the trained DTST and, for a fixed input, compare the indices produced by the Gumbel-Softmax relaxation in equation (2) with the indices produced by the exact hard top-$K$ rule in equation (1) under the same token scores; if the two index sets disagree on a substantial fraction of inputs, the model is not performing the token selection the paper describes, and one could retrain with straight-through hard selection to see whether the accuracy gains persist.

Watch

Extended reading notes

Core claim

The paper's central claim is that a View-Decoupled Transformer for aerial-ground person re-identification becomes more accurate when a Visual Token Selector (VTS) picks the $K=2$ most informative image tokens and discards the rest before final feature aggregation. The selector scores tokens with a softmax attention predictor, ranks them, and uses a Gumbel-Softmax relaxation of hard top-$K$ selection so the choice is trainable end to end. On the CARGO benchmark this configuration improves the VDT baseline by 3.13% in Rank-1, 1.18% in mAP, and 0.43% in mINP under the aerial-to-ground protocol, and by 1.60%, 1.51%, and 2.00% under the ALL protocol. On AG-ReID, Rank-1 rises by 0.57% for aerial-to-ground and 1.04% for ground-to-aerial. The paper interprets these gains as evidence that selecting fewer, identity-relevant tokens reduces redundancy and sharpens identity features.

Load-bearing premise

The whole accuracy story depends on equation (2) being a faithful differentiable stand-in for the hard top-$K$ selection in equation (1), because if the relaxation does not actually choose the top-$K$ tokens during training, the reported gains cannot be attributed to token selection.

Editorial extensions

If this is right

  • On CARGO's aerial-to-ground protocol, DTST reaches 50.63% Rank-1, 43.39% mAP, and 29.46% mINP, which beats VDT by 3.13, 1.18, and 0.43 points respectively.
  • On CARGO's ALL protocol, DTST exceeds VDT by 1.60 points Rank-1, 1.51 points mAP, and 2.00 points mINP.
  • On AG-ReID, DTST improves Rank-1 by 0.57 points for aerial-to-ground and 1.04 points for ground-to-aerial compared with VDT.
  • Ablations show that the best configuration uses the selector after the last layer with two attention heads and two selected tokens; raising the head count or selecting more than three tokens lowers accuracy.
  • Because only the top-$K$ tokens continue to the final aggregation stage, the method is intended to cut redundant computation, although the paper does not report measured inference time.

Reading between the lines

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

  • A testable extension the paper leaves implicit is whether the gains come from token pruning itself or from the extra parameters of the two-head selector; retraining with a fixed, non-selective two-token aggregation would isolate the mechanism.
  • The paper's efficiency claim is not backed by latency or FLOPs measurements, so a direct wall-clock comparison with the VDT baseline would determine whether the selector's overhead cancels the savings from discarding tokens.
  • Because the Visual Token Selector is architecture-agnostic, it could be grafted onto other view-heterogeneous retrieval pipelines or video models, but such transfer is an extension beyond the two benchmarks reported here.
  • One could test whether the Gumbel noise is necessary at all by replacing equation (2) with a straight-through hard top-$K$ estimator; equal accuracy would suggest the relaxation is incidental to the result.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 proposes a Dynamic Token Selective Transformer (DTST) for Aerial-Ground Person Re-identification (AGPReID), built on the View-Decoupled Transformer (VDT) baseline. The core novelty is a Visual Token Selector (VTS) that is claimed to rank image tokens by importance and select the top-K tokens in a differentiable manner via a Gumbel-Softmax relaxation. The authors report state-of-the-art results on the CARGO and AG-ReID datasets, with ablations over the number of selected tokens, attention heads, and token insertion positions.

Significance. The problem addressed is relevant and the authors choose appropriate public benchmarks (CARGO, AG-ReID) with standard evaluation metrics. The paper also provides a useful ablation perspective on token selection hyperparameters. However, the central methodological claim is not supported by the equations as written: the proposed differentiable top-K mechanism is not actually a top-K selector, and the baseline numbers are internally inconsistent across tables. If the method were correctly formulated and the experiments properly controlled, dynamic token selection for AGPReID could be a meaningful contribution; as presented, the contribution is not established.

major comments (4)
  1. [Section III-B, Eq. (1)] The scalar softmax in Eq. (1) is degenerate: for a scalar input x, softmax(x) = exp(x)/exp(x) = 1, so every token receives the same importance score and the TopK ranking is undefined. If the intended meaning is a softmax over the token dimension, the equation must explicitly normalize over j, e.g., s_i = exp(t_i^T Wq Wk^T t_i / sqrt(d)) / sum_j exp(...). As written, the proposed scoring mechanism cannot produce a meaningful ranking.
  2. [Section III-B, Eq. (2)] The Gumbel-Softmax expression is not a continuous relaxation of the TopK operation. As tau -> 0, the normalized exponential over (s_i + g_i)/tau converges to a one-hot distribution on a single argmax token, not to a selection of K tokens. For finite tau, it is a dense weighting that combines all M tokens, so no token is dropped, contradicting the claimed redundancy reduction. The implementation detail in Section IV-B that 'top two rated tokens' are selected is not derivable from Eq. (2); no repeated sampling without replacement, masking, or perturbed top-K optimizer is specified. Consequently, the described VTS does not implement the advertised top-K selection mechanism, and the reported accuracy gains cannot be attributed to it.
  3. [Section IV-C vs Section IV-D, Tables II and IV] The claimed improvement over the VDT baseline is internally inconsistent. In Table II, VDT achieves 47.50 Rank-1 and 42.21 mAP on the A↔G protocol of CARGO; in Table IV, model-a without the VTS achieves only 45.00 Rank-1 and 42.05 mAP on the same protocol. If model-a is intended to be the VDT baseline, the discrepancy changes the reported gains from 3.13% to 5.63% in Rank-1 and from 1.18% to 1.34% in mAP. The source of this discrepancy is unexplained, so the ablation in Table IV does not cleanly isolate the contribution of the VTS.
  4. [Section IV-D, Table V] The final model configuration (2 attention heads, top-2 selected tokens, last-layer insertion) is chosen by evaluating configurations on the CARGO A↔G protocol, which is the same protocol used for the state-of-the-art comparison in Table II. This means the reported SOTA result is the best result from a search over the test evaluation protocol, which inflates the expected performance and makes the comparison to prior methods optimistic. A validation split or repeated cross-validation should be used for hyperparameter selection before reporting SOTA numbers.
minor comments (5)
  1. [Section IV-D, Table V] The table header contains typos: 'mMAP' should be 'mAP' and 'blod' should be 'bold'; also, the text says 'conducted on the on the CARGO dataset', with a duplicated 'on the'.
  2. [Section II-A and Table I] The AG-ReID dataset is attributed to reference [12] in the Related Work but to reference [19] in Table I and Section IV-A; the authors should clarify which version of the dataset is used and cite consistently.
  3. [Section II-B] The sentence 'Building on the principles of differentiable top-K selection [18]' cites reference [18] (TS2-Net), but the differentiable perturbed-optimizer method described in the text is reference [17] (Berthet et al.); the citation appears incorrect.
  4. [Section IV-D] The phrase 'insertion position of VST' should read 'insertion position of VTS' for consistency with the abbreviated module name.
  5. [Tables II and III] The paper does not report whether results are averaged over multiple runs or whether evaluation is performed on a single test protocol; reporting run-to-run variance or at least noting single-run results would improve reliability.

Circularity Check

1 steps flagged · score 6.0 of 10

CARGO headline mAP is the best of a test-protocol hyperparameter search, so the reported state-of-the-art gain is a selected statistic rather than a holdout prediction.

  1. fitted input called prediction [Section IV-D (Ablation Study), Table V and Table II, CARGO A↔G protocol]
    "Notably, all ablation experiments are conducted on the on the CARGO dataset. ... model-6 (Ours) 2 2 last layer 50.63 43.39 29.46 ... DTST (Ours) 64.42 55.73 41.92 78.57 72.40 62.10 80.00 63.31 44.67 50.63 43.39 29.46"

    The final model configuration (2 attention heads, 2 selected tokens, insertion at the last layer) is chosen because it achieves the best Rank-1/mAP on the CARGO A↔G protocol in the ablation grid. Table V's model-6 row is then reported verbatim as 'DTST (Ours)' in Table II with the same A↔G numbers. The abstract's headline '+1.18% mAP over the second place' is therefore the maximum of a hyperparameter search evaluated on the very test protocol that is later reported as the result. On this protocol the headline performance is a fitted/selected statistic by construction, not an independent prediction.

full rationale

Aside from the test-protocol hyperparameter selection, I find no load-bearing circularity. The VDT baseline [15] is an external method, and the differentiable top-K references [16]-[18] are not self-citations, so there is no self-citation chain forcing the conclusion. The Gumbel-Softmax equation in Section III-B is not a faithful continuous relaxation of Top-K (it is a dense softmax that collapses to a single argmax as temperature goes to zero, not to K discrete tokens), but that is a technical correctness and reproducibility gap rather than an equivalence between inputs and outputs, so under the hard rules I do not count it as circularity. The AG-ReID results in Table III are on a dataset not used in the ablation search and provide some independent support for the method, which is why the score is 6 (partial) rather than higher. The CARGO A↔G headline claim itself, however, reduces to the best configuration from a search on that same protocol, making that specific state-of-the-art claim partially circular by construction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The central mechanism rests on an unproven equivalence between Gumbel-Softmax and top-K selection, on the availability of accurate view labels, on the transferability of ImageNet pretraining, and on three hyperparameters chosen by test-set ablation. None of these are independently verified in the paper.

free parameters (4)
  • K (number of selected tokens) = 2
    Chosen based on ablation on the test protocol A↔G (Table V); the paper selects top-2 tokens as best.
  • number of attention heads in VTS = 2
    Selected after testing 2, 4, 8 heads; 2 heads gave best Rank1/mAP on test protocol (Table V).
  • token insertion position = last layer
    Selected as best performing insertion position after comparing with second-to-last layer (Table V).
  • Gumbel temperature tau = not reported
    Appears in the Gumbel-Softmax equation (Section III-B) but no value or schedule is given.
assumptions (4)
  • domain assumption Attention scores computed as t_i^T W_q W_k^T t_i reflect token importance for identity recognition
    This is the basis for ranking tokens in the Visual Token Selector; it is assumed in Section III-B.
  • ad hoc to paper Gumbel-Softmax is a valid continuous relaxation of Top-K selection
    Invoked in Section III-B, Eq. (2); however Gumbel-Softmax is a relaxation of argmax/categorical sampling, not top-K, so this assumption is likely false.
  • domain assumption View labels are available and correctly assigned for training
    The VDT uses view labels and an orthogonal loss; the paper assumes these labels are reliable (Section III-A).
  • domain assumption ImageNet-pretrained ViT-Base provides a good starting point for AGPReID
    Used in implementation details (Section IV-B); transferability is assumed.
invented entities (1)
  • Visual Token Selector (VTS) module
    purpose: Selects top-K informative tokens to reduce redundancy and focus on discriminative regions
    A new architectural module introduced in Section III-B; its only claimed evidence is benchmark performance, no external falsifiable prediction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamic Token Selection for Aerial-Ground Person Re-Identification." pith.science (2026). https://pith.science/paper/EKXFJO3A

@misc{pith2026241200433,
  author       = {Pith},
  title        = {Pith review of: Dynamic Token Selection for Aerial-Ground Person Re-Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EKXFJO3A}},
  note         = {Machine review of arXiv:2412.00433}
}
read the original abstract

Aerial-Ground Person Re-identification (AGPReID) holds significant practical value but faces unique challenges due to pronounced variations in viewing angles, lighting conditions, and background interference. Traditional methods, often involving a global analysis of the entire image, frequently lead to inefficiencies and susceptibility to irrelevant data. In this paper, we propose a novel Dynamic Token Selective Transformer (DTST) tailored for AGPReID, which dynamically selects pivotal tokens to concentrate on pertinent regions. Specifically, we segment the input image into multiple tokens, with each token representing a unique region or feature within the image. Using a Top-k strategy, we extract the k most significant tokens that contain vital information essential for identity recognition. Subsequently, an attention mechanism is employed to discern interrelations among diverse tokens, thereby enhancing the representation of identity features. Extensive experiments on benchmark datasets showcases the superiority of our method over existing works. Notably, on the CARGO dataset, our proposed method gains 1.18% mAP improvements when compared to the second place. In addition, we comprehensively analyze the impact of different numbers of tokens, token insertion positions, and numbers of heads on model performance.

Figures

Figures reproduced from arXiv: 2412.00433 by the authors.

Figure 1
Figure 1. A straightforward description of Aerial-Ground Per [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the proposed Dynamic Token Selective Transformer (DTST) framework. The framework incorporates [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The Illustration of Visual Token Selector (VTS). The process involves selecting the Top-K informative tokens from the [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗

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. HiHR: Hierarchical Hyperbolic Representation for Aerial-Ground Person Re-Identification

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Hierarchical hyperbolic embeddings with text-guided multi-granularity fusion improve aerial-ground person re-identification by keeping both view-invariant identity and view-specific cues.

Reference graph

Works this paper leans on

26 extracted references · 24 canonical work pages · cited by 1 Pith paper

  1. [1]

    Dissecting person re-identification from the viewpoint of viewpoint,

    Xiaoxiao Sun and Liang Zheng, “Dissecting person re-identification from the viewpoint of viewpoint,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 608– 617

    work page 2019

  2. [2]

    Self- supervised image-specific prototype exploration for weakly supervised semantic segmentation,

    Qi Chen, Lingxiao Yang, Jian-Huang Lai, and Xiaohua Xie, “Self- supervised image-specific prototype exploration for weakly supervised semantic segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2022, pp. 4288–4298

    work page 2022

  3. [3]

    Spike count maximization for neuromorphic vision recognition.,

    Jianxiong Tang, Jian-Huang Lai, Xiaohua Xie, and Lingxiao Yang, “Spike count maximization for neuromorphic vision recognition.,” in IJCAI, 2023, pp. 4253–4261

    work page 2023

  4. [4]

    Pose guided person image generation via dual-task correlation and affinity learning,

    Pengze Zhang, Lingxiao Yang, Xiaohua Xie, and Jianhuang Lai, “Pose guided person image generation via dual-task correlation and affinity learning,” IEEE Transactions on Visualization and Computer Graphics , 2023

    work page 2023

  5. [5]

    Region-based online selective examination for weakly supervised se- mantic segmentation,

    Qi Chen, Yun Chen, Yuheng Huang, Xiaohua Xie, and Lingxiao Yang, “Region-based online selective examination for weakly supervised se- mantic segmentation,” Information Fusion, vol. 107, pp. 102311, 2024

    work page 2024

  6. [6]

    Camera-driven representation learning for unsupervised domain adaptive person re-identification,

    Geon Lee, Sanghoon Lee, Dohyung Kim, Younghoon Shin, Yongsang Yoon, and Bumsub Ham, “Camera-driven representation learning for unsupervised domain adaptive person re-identification,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 11453–11462

    work page 2023

  7. [7]

    Weperson: Learning a generalized re- identification model from all-weather virtual data,

    He Li, Mang Ye, and Bo Du, “Weperson: Learning a generalized re- identification model from all-weather virtual data,” in Proceedings of the 29th ACM international conference on multimedia , 2021, pp. 3115– 3123

    work page 2021

  8. [8]

    Uncertainty modeling with second-order transformer for group re- identification,

    Quan Zhang, Jian-Huang Lai, Zhanxiang Feng, and Xiaohua Xie, “Uncertainty modeling with second-order transformer for group re- identification,” in Proceedings of the AAAI conference on artificial intelligence, 2022, vol. 36, pp. 3318–3325

    work page 2022

Show all 26 references
  1. [9]

    Separable spatial-temporal residual graph for cloth-changing group re-identification,

    Quan Zhang, Jianhuang Lai, Xiaohua Xie, Xiaofeng Jin, and Sien Huang, “Separable spatial-temporal residual graph for cloth-changing group re-identification,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  2. [10]

    Ac2as: Activation consistency coupled ann-snn framework for fast and memory-efficient snn training,

    Jianxiong Tang, Jian-Huang Lai, Xiaohua Xie, Lingxiao Yang, and Wei- Shi Zheng, “Ac2as: Activation consistency coupled ann-snn framework for fast and memory-efficient snn training,” Pattern Recognition, vol. 144, pp. 109826, 2023

    2023

  3. [11]

    Salient part-aligned and keypoint disentangling transformer for person re- identification in aerial imagery,

    Junyang Qiu, Zhanxiang Feng, Lei Wang, and Jianhuang Lai, “Salient part-aligned and keypoint disentangling transformer for person re- identification in aerial imagery,” in 2024 IEEE International Conference on Multimedia and Expo (ICME) . IEEE, 2024, pp. 1–6

    2024

  4. [12]

    Aerial-ground person re-id,

    Huy Nguyen, Kien Nguyen, Sridha Sridharan, and Clinton Fookes, “Aerial-ground person re-id,” in 2023 IEEE International Conference on Multimedia and Expo (ICME) . IEEE, 2023, pp. 2585–2590

    2023

  5. [13]

    Scalable person re-identification: A benchmark,

    Liang Zheng, Liyue Shen, Lu Tian, Shengjin Wang, Jingdong Wang, and Qi Tian, “Scalable person re-identification: A benchmark,” in Proceedings of the IEEE international conference on computer vision , 2015, pp. 1116–1124

    2015

  6. [14]

    Person transfer gan to bridge domain gap for person re-identification,

    Longhui Wei, Shiliang Zhang, Wen Gao, and Qi Tian, “Person transfer gan to bridge domain gap for person re-identification,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2018, pp. 79–88

    2018

  7. [15]

    View-decoupled transformer for person re-identification under aerial-ground camera network,

    Quan Zhang, Lei Wang, Vishal M Patel, Xiaohua Xie, and Jianhaung Lai, “View-decoupled transformer for person re-identification under aerial-ground camera network,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 22000–22009

    2024

  8. [16]

    Efficient video transformers with spatial-temporal token selection,

    Junke Wang, Xitong Yang, Hengduo Li, Li Liu, Zuxuan Wu, and Yu- Gang Jiang, “Efficient video transformers with spatial-temporal token selection,” in European Conference on Computer Vision. Springer, 2022, pp. 69–86

    2022

  9. [17]

    Learning with differentiable pertubed optimizers,

    Quentin Berthet, Mathieu Blondel, Olivier Teboul, Marco Cuturi, Jean- Philippe Vert, and Francis Bach, “Learning with differentiable pertubed optimizers,” Advances in neural information processing systems , vol. 33, pp. 9508–9519, 2020

    2020

  10. [18]

    Ts2- net: Token shift and selection transformer for text-video retrieval,

    Yuqi Liu, Pengfei Xiong, Luhui Xu, Shengming Cao, and Qin Jin, “Ts2- net: Token shift and selection transformer for text-video retrieval,” in European conference on computer vision . Springer, 2022, pp. 319–335

    2022

  11. [19]

    Ag- reid.v2: Bridging aerial and ground views for person re-identification,

    Huy Nguyen, Kien Nguyen, Sridha Sridharan, and Clinton Fookes, “Ag- reid.v2: Bridging aerial and ground views for person re-identification,” IEEE Transactions on Information Forensics and Security , vol. 19, pp. 2896–2908, 2024

    2024

  12. [20]

    Fastreid: A pytorch toolbox for general instance re-identification,

    Lingxiao He, Xingyu Liao, Wu Liu, Xinchen Liu, Peng Cheng, and Tao Mei, “Fastreid: A pytorch toolbox for general instance re-identification,” in ACM Int. Conf. Multimedia , 2023, pp. 9664–9667

    2023

  13. [21]

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

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

    2019

  14. [22]

    Learning discriminative features with multiple granularities for person re-identification,

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

    2018

  15. [23]

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

    Mang Ye, Jianbing Shen, Gaojie Lin, Tao Xiang, Ling Shao, and Steven C. H. Hoi, “Deep learning for person re-identification: A survey and outlook,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 44, no. 6, pp. 2872–2893, 2022

    2022

  16. [24]

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

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

    2010 arXiv

  17. [25]

    Learning generalisable omni-scale representations for person re- identification,

    Kaiyang Zhou, Yongxin Yang, Andrea Cavallaro, and Tao Xiang, “Learning generalisable omni-scale representations for person re- identification,” IEEE transactions on pattern analysis and machine intelligence, vol. 44, no. 9, pp. 5056–5069, 2021

    2021

  18. [26]

    An image is worth more than 16x16 patches: Exploring transformers on individual pixels,

    Duy-Kien Nguyen, Mahmoud Assran, Unnat Jain, Martin R Oswald, Cees GM Snoek, and Xinlei Chen, “An image is worth more than 16x16 patches: Exploring transformers on individual pixels,” arXiv preprint arXiv:2406.09415, 2024

    2024 arXiv

Pith tools

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