Pith. sign in

REVIEW 4 major objections 5 minor 71 references

Video-Level Language-Driven Video-Based Visible-Infrared Person Re-Identification

T0 review · 4 major / 5 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 VLD uses CLIP-generated video-level language prompts to bridge the visible-infrared gap in video person re-identification, achieving state-of-the-art results on two benchmarks with modest extra cost.

desk verdict Solid engineering with real benchmark gains, but the 'language-driven' mechanism is asserted more than shown; the paper deserves peer review, not a desk reject. read the letter →

arxiv 2506.02439 v1 pith:OXQVQFZS submitted 2025-06-03 cs.CV

Shuang Li , Jiaxu Leng , Changjiang Kuang , Mingpi Tan , Xinbo Gao This is my paper
classification cs.CV
keywords VVI-ReIDvisible-infraredpersonre-identificationCLIPtextpromptingspatiotemporalvideo-basedmodalityalignmentcross-modalretrieval
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 tries to establish that language descriptions of pedestrian sequences can serve as a modality-shared bridge for video-based visible-infrared person re-identification (VVI-ReID), a task where matching must survive the large spectral gap between daytime RGB and nighttime infrared footage. The authors propose a framework, VLD, that adapts the CLIP vision-language model to this video task without requiring any manually written text annotations. VLD jointly fine-tunes the CLIP visual encoder and a learnable prompt generator so that the frozen text encoder produces identity-level prototypes that classify both visible and infrared video features, while a lightweight spatial-temporal prompting module injects motion and cross-frame information into the model. If the paper's claims hold, language-driven alignment plus spatiotemporal prompting is a practical recipe for VVI-ReID that also transfers better across datasets and trains in about two hours.

What carries the argument

The load-bearing object is the invariant-modality language prompting (IMLP) mechanism, which treats the frozen CLIP text encoder plus learnable identity-level text prompts as a fixed classifier: text features for all identities are computed once per iteration and used as prototypes in a visual-to-text cross-entropy loss against both visible and infrared sequence features. The second mechanism is the spatial-temporal hub (STH), a learnable tensor of shape $T\times T\times D$ that is concatenated with patch tokens inside selected ViT layers and alternately transposed, letting multihead attention aggregate spatial information from all frames and then diffuse it back to each frame's [CLS] token, with spatial-temporal aggregation (STA) adding a direct attention path and identity-level losses to keep the hub identity-relevant.

What would settle it

Train VLD with the text encoder unfrozen (or replace the frozen-text-prototype classifier with a learned linear classifier of the same capacity) and compare on HITSZ-VCM. If performance does not drop substantially, the frozen-text-prototype alignment is not the cause of the improvement. A direct check of the alignment mechanism is to measure the average distance between visible and infrared features of the same identity in the shared embedding space during training: if IMLP does not reduce this gap relative to the baseline without the visual-to-text loss, the language bridge is not actually aligning modalities.

Watch

Extended reading notes

Core claim

The central claim is that modality-invariant sequence-level pedestrian features can be learned by using video-level language prompts as the alignment mechanism, rather than relying only on one-hot identity labels. The proposed VLD framework contains two modules: invariant-modality language prompting (IMLP), which treats the combination of a frozen CLIP text encoder and learnable identity-specific text prompts as an identity classifier and aligns both visible and infrared visual features to those text prototypes through a visual-to-text loss; and spatial-temporal prompting (STP), which injects a learnable spatial-temporal hub tensor into intermediate ViT layers, alternately transposing it to aggregate and diffuse spatial and temporal information across frames, plus a spatial-temporal aggregation module with dedicated attention and identity losses. On the HITSZ-VCM dataset, VLD improves Rank-1 accuracy by 7.3% and mAP by 7.6% for infrared-to-visible retrieval and by 10.4% and 9.3% for visible-to-infrared retrieval over the baseline, and it also sets new state-of-the-art results on BUPTCampus while adding only 2.39M parameters and 0.12G FLOPs.

Load-bearing premise

The method assumes that jointly fine-tuning the CLIP visual encoder and the prompt learner preserves enough of CLIP's original multimodal embedding space that the frozen text encoder's prototypes still act as valid classifiers for both visible and infrared video features. If the text prototypes and visual features drift apart during training, the visual-to-text loss degenerates into a linear classifier and the claimed language-driven modality alignment has no mechanism.

Editorial extensions

If this is right

  • If VLD is correct, language-driven alignment is a viable alternative to adversarial or GAN-based modality alignment for VVI-ReID, with lower training cost and better cross-dataset transfer.
  • The framework suggests that frozen text encoders from CLIP can serve as reusable semantic classifiers for new visual tasks even when the visual encoder is fine-tuned, as long as prompts are learned jointly.
  • The STP mechanism shows that spatiotemporal information can be added to a ViT with negligible parameter overhead, potentially replacing heavier post-encoders like LSTM or transformer aggregates in video ReID.
  • VLD's cross-dataset gains imply that language-aligned features generalize better across domains and cameras, which is directly relevant to day-and-night surveillance deployment.
  • The reported two-hour training time and small overhead indicate the method is practical for large-scale or incremental deployment.

Reading between the lines

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

  • The idea of using frozen-text-encoder prototypes as a classifier could generalize beyond person ReID to other cross-modal retrieval tasks where text annotations are unavailable, such as vehicle or animal re-identification from thermal and RGB cameras.
  • The STH transposition trick is a generic way to let a ViT exchange information across a batch of frames; it could be lifted to other video understanding tasks that need lightweight temporal modeling without architectural changes.
  • A testable extension would be to evaluate VLD with different text template languages (or multilingual templates) to see whether the language channel is truly semantic or just a flexible learned embedding.
  • The authors' limitation note about fixed-length temporal sampling suggests that an adaptive-length version, perhaps with a temporal attention over variable tracklet lengths, would be a natural next step that could further improve robustness.
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 proposes VLD, a framework for video-based visible-infrared person re-identification (VVI-ReID) built on CLIP. It comprises two modules: IMLP, which treats learnable identity-level text prompts processed by a frozen CLIP text encoder as an identity classifier for sequence-level visual features via a visual-to-text loss (Eq. 5), and STP, which inserts a learnable spatial-temporal hub (STH) into intermediate ViT layers and adds an aggregation module (STA) with identity and triplet losses. The framework is evaluated on HITSZ-VCM and BUPTCampus, reporting state-of-the-art results with modest added parameters and FLOPs, plus cross-dataset transfer experiments.

Significance. If confirmed, this would be the first CLIP-based VVI-ReID method and introduces an efficient spatiotemporal prompting mechanism. The reported gains over strong baselines are substantial, and the computational overhead is small. The paper includes ablations, cross-dataset tests, visualization, and a complexity analysis, which are useful. However, the central language-driven mechanism is not convincingly established, and the experimental protocol has weaknesses that need to be addressed before the state-of-the-art claim can be accepted. The cross-dataset generalization results (Table VIII) are a notable strength, as they go beyond single-dataset evaluation.

major comments (4)
  1. [Section III-C, Eq. (5)]
  2. [Section IV-B, Implementation Details]
  3. [Section IV-D (Tables VI and VII) and Section IV-E (Fig. 6)]
  4. [Table IV]
minor comments (5)
  1. [Section IV-C, Evaluation on HITSZ-VCM]
  2. [Tables I and II]
  3. [Equation (5)]
  4. [Section III-D, Spatial-Temporal Hub]
  5. [Abstract and Conclusion]

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: VLD's SOTA claims are empirical evaluations on external benchmarks; IMLP's training-with-identity-labels concern is a mechanism/anchoring issue, not a definitional reduction.

full rationale

The paper's central claims are empirical benchmark results, not derivations from fitted quantities. IMLP (Sec. III-C) treats the frozen text encoder plus learnable prompts as an identity classifier; Eq. 5 is a softmax over text prototypes, and those prototypes are trained with the same identity labels as the visual classifier. This is a legitimate concern about whether the module is truly 'language-driven' or just an additional linear classifier, but it is not circular: the paper does not claim the text supervision is independent of the labels, and it does not define the language bridge by Eq. 5's value. The SOTA numbers on HITSZ-VCM and BUPTCampus, and the cross-dataset results in Table VIII, are evaluated on held-out identities and external datasets, so they do not reduce by construction to a fitted parameter. The claim that a frozen text encoder keeps visual features in CLIP's original multimodal space is at most an unsupported inference (freezing one side does not fix the other), which is a correctness/mechanism issue, not a self-definitional reduction. The prompt-template ablation in Table VII selects a template based on test performance, which is a test-set selection concern rather than circularity. No load-bearing self-citation chain or imported uniqueness theorem is present. Mild burden, no circularity.

Assumptions & free parameters 6 free parameters · 3 assumptions · 1 invented entities

All learned components are trained end-to-end. The central claim therefore rests on ordinary supervised training assumptions: the CLIP text encoder retains discriminative meaning after prompt learning, a single prompt can represent one identity across modalities, and the STH attention mechanism actually transfers spatiotemporal information. These are design assumptions backed only by internal ablations, with no external or machine-checked handles, and all hyperparameters are tuned on the same benchmarks used for the final SOTA claim.

free parameters (6)
  • lambda_1 (L_v2t weight) = 0.08
    Selected by a parameter scan on the HITSZ-VCM test protocol, Figure 6(a).
  • lambda_2 (L_H_id weight) = 0.4
    Selected by a parameter scan on the HITSZ-VCM test protocol, Figure 6(b).
  • lambda_3 (L_H_wrt weight) = 1.0
    Selected by a parameter scan on the HITSZ-VCM test protocol, Figure 6(c).
  • STH insertion layer = 9
    Chosen from an ablation on the test set, Table VI; the layer number is treated as a hand-tuned design choice rather than derived.
  • Prompt template = Template 4: 'A [X]1... person observed in both day and night conditions'
    Selected from Table VII based on test-set Rank-1 and mAP.
  • Number of learnable prompt tokens M = 4
    Set in Section IV-B without a reported sensitivity analysis.
assumptions (3)
  • domain assumption The frozen CLIP text encoder, combined with learnable prompt tokens, yields identity-discriminative prototypes in a shared embedding space even when the visual encoder is jointly fine-tuned.
    Invoked in Section III-C during classifier construction and visual classification, Eq. 5; there is no independent check that the learned prompts retain semantic interpretability.
  • domain assumption The infrared and visible frames of the same identity share enough high-level visual semantics for a single video-level text prompt to represent both.
    This is the foundational motivation in Section I and Figure 1; if false, IMLP cannot align modalities through text.
  • domain assumption Concatenating an extra learned tensor H to patch tokens and alternately transposing H propagates spatiotemporal information through the ViT's multihead attention into the [CLS] tokens.
    Assumed in Section III-D, Eq. 6 to Eq. 9; there is no theoretical proof, only ablation evidence.
invented entities (1)
  • Spatial-temporal hub H
    purpose: A learned T x T x D tensor injected into ViT layers to aggregate and diffuse spatiotemporal information across frames.
    Its role is validated only by ablation gains on the same benchmarks where all other components are trained and tuned; there is no external or falsifiable evidence that H itself captures identity-relevant temporal structure rather than acting as extra model capacity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Video-Level Language-Driven Video-Based Visible-Infrared Person Re-Identification." pith.science (2026). https://pith.science/paper/OXQVQFZS

@misc{pith2026250602439,
  author       = {Pith},
  title        = {Pith review of: Video-Level Language-Driven Video-Based Visible-Infrared Person Re-Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OXQVQFZS}},
  note         = {Machine review of arXiv:2506.02439}
}
read the original abstract

Video-based Visible-Infrared Person Re-Identification (VVI-ReID) aims to match pedestrian sequences across modalities by extracting modality-invariant sequence-level features. As a high-level semantic representation, language provides a consistent description of pedestrian characteristics in both infrared and visible modalities. Leveraging the Contrastive Language-Image Pre-training (CLIP) model to generate video-level language prompts and guide the learning of modality-invariant sequence-level features is theoretically feasible. However, the challenge of generating and utilizing modality-shared video-level language prompts to address modality gaps remains a critical problem. To address this problem, we propose a simple yet powerful framework, video-level language-driven VVI-ReID (VLD), which consists of two core modules: invariant-modality language prompting (IMLP) and spatial-temporal prompting (STP). IMLP employs a joint fine-tuning strategy for the visual encoder and the prompt learner to effectively generate modality-shared text prompts and align them with visual features from different modalities in CLIP's multimodal space, thereby mitigating modality differences. Additionally, STP models spatiotemporal information through two submodules, the spatial-temporal hub (STH) and spatial-temporal aggregation (STA), which further enhance IMLP by incorporating spatiotemporal information into text prompts. The STH aggregates and diffuses spatiotemporal information into the [CLS] token of each frame across the vision transformer (ViT) layers, whereas STA introduces dedicated identity-level loss and specialized multihead attention to ensure that the STH focuses on identity-relevant spatiotemporal feature aggregation. The VLD framework achieves state-of-the-art results on two VVI-ReID benchmarks. The code will be released at https://github.com/Visuang/VLD.

Figures

Figures reproduced from arXiv: 2506.02439 by the authors.

Figure 1
Figure 1. Modality-shared video-level language effectively describes pedestrian [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The core motivations of this paper: (a) CLIP-ReID struggles to [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The proposed VLD framework consists of two modules: invariant-modality language prompting (IMLP) and spatial-temporal prompting (STP). IMLP [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: An illustration of the proposed STA, which employs specialized [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Results of Rank-1 and mAP with different values of [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Visualization of intra-class and inter-class distance distributions. (a) [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 9
Figure 9. Figure 9: Visualization of pedestrian search results, where B/L represents the baseline method. For each query pedestrian sequence, the top 6 most similar [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 63 canonical work pages

  1. [1]

    Appearance- preserving 3d convolution for video-based person re-identification,

    X. Gu, H. Chang, B. Ma, H. Zhang, and X. Chen, “Appearance- preserving 3d convolution for video-based person re-identification,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16. Springer, 2020, pp. 228–243

    work page 2020

  2. [2]

    Watching you: Global- guided reciprocal learning for video-based person re-identification,

    X. Liu, P. Zhang, C. Yu, H. Lu, and X. Yang, “Watching you: Global- guided reciprocal learning for video-based person re-identification,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 13 334–13 343

    work page 2021

  3. [3]

    Pyramid spatial-temporal aggregation for video-based person re-identification,

    Y . Wang, P. Zhang, S. Gao, X. Geng, H. Lu, and D. Wang, “Pyramid spatial-temporal aggregation for video-based person re-identification,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 12 026–12 035

    work page 2021

  4. [4]

    Salient-to-broad transition for video person re-identification,

    S. Bai, B. Ma, H. Chang, R. Huang, and X. Chen, “Salient-to-broad transition for video person re-identification,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7339–7348

    2022

  5. [5]

    Deeply coupled convolution– transformer with spatial–temporal complementary learning for video- based person re-identification,

    X. Liu, C. Yu, P. Zhang, and H. Lu, “Deeply coupled convolution– transformer with spatial–temporal complementary learning for video- based person re-identification,”IEEE Transactions on Neural Networks and Learning Systems, 2023

    work page 2023

  6. [6]

    A video is worth three views: Trigeminal transformers for video-based person re- identification,

    X. Liu, P. Zhang, C. Yu, X. Qian, X. Yang, and H. Lu, “A video is worth three views: Trigeminal transformers for video-based person re- identification,”IEEE Transactions on Intelligent Transportation Systems, 2024

    work page 2024

  7. [7]

    Tf-clip: Learning text- free clip for video-based person re-identification,

    C. Yu, X. Liu, Y . Wang, P. Zhang, and H. Lu, “Tf-clip: Learning text- free clip for video-based person re-identification,” inAAAI Conference on Artificial Intelligence, AAAI, vol. 38, no. 7, 2024, pp. 6764–6772

    work page 2024

  8. [8]

    Body part-level domain alignment for domain-adaptive person re-identification with transformer framework,

    Y . Wang, G. Qi, S. Li, Y . Chai, and H. Li, “Body part-level domain alignment for domain-adaptive person re-identification with transformer framework,”IEEE Transactions on Information Forensics and Security, vol. 17, pp. 3321–3334, 2022

    work page 2022

Show all 71 references
  1. [9]

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

    M. Ye, J. Shen, G. Lin, T. Xiang, L. Shao, and S. C. Hoi, “Deep learning for person re-identification: A survey and outlook,”IEEE transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 6, pp. 2872– 2893, 2021

    2021

  2. [10]

    Partial person re-identification,

    W.-S. Zheng, X. Li, T. Xiang, S. Liao, J. Lai, and S. Gong, “Partial person re-identification,” inProceedings of the IEEE International Conference on Computer Vision, 2015, pp. 4678–4686

    2015

  3. [11]

    Harmonious attention network for person re-identification,

    W. Li, X. Zhu, and S. Gong, “Harmonious attention network for person re-identification,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 2285–2294

    2018

  4. [12]

    A survey of open-world person re- identification,

    Q. Leng, M. Ye, and Q. Tian, “A survey of open-world person re- identification,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 30, no. 4, pp. 1092–1108, 2019

    2019

  5. [13]

    Dynamic tri-level relation mining with attentive graph for visible infrared re-identification,

    M. Ye, C. Chen, J. Shen, and L. Shao, “Dynamic tri-level relation mining with attentive graph for visible infrared re-identification,”IEEE Transactions on Information Forensics and Security, vol. 17, pp. 386– 398, 2021

    2021

  6. [14]

    Dual consistency-constrained learning for unsupervised visible-infrared person re-identification,

    B. Yang, J. Chen, C. Chen, and M. Ye, “Dual consistency-constrained learning for unsupervised visible-infrared person re-identification,”IEEE Transactions on Information Forensics and Security, 2023

    2023

  7. [15]

    Securereid: Privacy- preserving anonymization for person re-identification,

    M. Ye, W. Shen, J. Zhang, Y . Yang, and B. Du, “Securereid: Privacy- preserving anonymization for person re-identification,”IEEE Transac- tions on Information Forensics and Security, 2024

    2024

  8. [16]

    Performance measures and a data set for multi-target, multi-camera tracking,

    E. Ristani, F. Solera, R. Zou, R. Cucchiara, and C. Tomasi, “Performance measures and a data set for multi-target, multi-camera tracking,” in European conference on computer vision. Springer, 2016, pp. 17–35

    2016

  9. [17]

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

    L. Wei, S. Zhang, W. Gao, and Q. Tian, “Person transfer gan to bridge domain gap for person re-identification,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 79– 88

    2018

  10. [18]

    Mars: A video benchmark for large-scale person re-identification,

    L. Zheng, Z. Bie, Y . Sun, J. Wang, C. Su, S. Wang, and Q. Tian, “Mars: A video benchmark for large-scale person re-identification,” inCom- puter Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part VI 14. Springer, 201...

    2016

  11. [19]

    Global-local temporal representations for video person re-identification,

    J. Li, J. Wang, Q. Tian, W. Gao, and S. Zhang, “Global-local temporal representations for video person re-identification,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 3958–3967

    2019

  12. [20]

    Exploit the unknown gradually: One-shot video-based person re-identification by stepwise learning,

    Y . Wu, Y . Lin, X. Dong, Y . Yan, W. Ouyang, and Y . Yang, “Exploit the unknown gradually: One-shot video-based person re-identification by stepwise learning,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 5177–5186

    2018

  13. [21]

    Person re-identification by video ranking,

    T. Wang, S. Gong, X. Zhu, and S. Wang, “Person re-identification by video ranking,” inComputer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part IV 13. Springer, 2014, pp. 688–703

    2014

  14. [22]

    Person re- identification by descriptive and discriminative classification,

    M. Hirzer, C. Beleznai, P. M. Roth, and H. Bischof, “Person re- identification by descriptive and discriminative classification,” inImage Analysis: 17th Scandinavian Conference, SCIA 2011, Ystad, Sweden, May 2011. Proceedings 17. Springer, 2011, pp. 91–102

    2011

  15. [23]

    Learning modal-invariant and temporal-memory for video-based visible-infrared person re-identification,

    X. Lin, J. Li, Z. Ma, H. Li, S. Li, K. Xu, G. Lu, and D. Zhang, “Learning modal-invariant and temporal-memory for video-based visible-infrared person re-identification,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 20 973–20 982

    2022

  16. [24]

    Adversarial self-attack defense and spatial-temporal relation mining for visible-infrared video person re-identification,

    H. Li, L. Xu, Y . Zhang, D. Tao, and Z. Yu, “Adversarial self-attack defense and spatial-temporal relation mining for visible-infrared video person re-identification,”arXiv preprint arXiv:2307.03903, 2023

    2023 arXiv

  17. [25]

    Video-based visible- infrared person re-identification via style disturbance defense and dual interaction,

    C. Zhou, J. Li, H. Li, G. Lu, Y . Xu, and M. Zhang, “Video-based visible- infrared person re-identification via style disturbance defense and dual interaction,” inProceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 46–55

    2023

  18. [26]

    Video-based visible-infrared person re-identification with auxiliary samples,

    Y . Du, C. Lei, Z. Zhao, Y . Dong, and F. Su, “Video-based visible-infrared person re-identification with auxiliary samples,”IEEE Transactions on Information Forensics and Security, vol. 19, pp. 1313–1325, 2023

    2023

  19. [27]

    Intermediary-guided bidi- rectional spatial-temporal aggregation network for video-based visible- infrared person re-identification,

    H. Li, M. Liu, Z. Hu, F. Nie, and Z. Yu, “Intermediary-guided bidi- rectional spatial-temporal aggregation network for video-based visible- infrared person re-identification,”IEEE Transactions on Circuits and Systems for Video Technology, 2023

    2023

  20. [28]

    Clip-reid: exploiting vision-language model for image re-identification without concrete text labels,

    S. Li, L. Sun, and Q. Li, “Clip-reid: exploiting vision-language model for image re-identification without concrete text labels,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 1, 2023, pp. 1405–1413

    2023

  21. [29]

    Cross-modality spatial-temporal transformer for video- based visible-infrared person re-identification,

    Y . Feng, F. Chen, J. Yu, Y . Ji, F. Wu, T. Liu, S. Liu, X.-Y . Jing, and J. Luo, “Cross-modality spatial-temporal transformer for video- based visible-infrared person re-identification,”IEEE Transactions on Multimedia, 2024

    2024

  22. [30]

    Prompt switch: Efficient clip adaptation for text-video retrieval,

    C. Deng, Q. Chen, P. Qin, D. Chen, and Q. Wu, “Prompt switch: Efficient clip adaptation for text-video retrieval,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 15 648–15 658

    2023

  23. [31]

    Idea: Inverted text with cooper- ative deformable aggregation for multi-modal object re-identification,

    Y . Wang, Y . Lv, P. Zhang, and H. Lu, “Idea: Inverted text with cooper- ative deformable aggregation for multi-modal object re-identification,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025

    2025

  24. [32]

    Decoupled feature-based mixture of experts for multi-modal object re-identification,

    Y . Wang, Y . Liu, A. Zheng, and P. Zhang, “Decoupled feature-based mixture of experts for multi-modal object re-identification,” inProceed- ings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 8, 2025, pp. 8141–8149

    2025

  25. [33]

    Mam- bapro: Multi-modal object re-identification with mamba aggregation and synergistic prompt,

    Y . Wang, X. Liu, T. Yan, Y . Liu, A. Zheng, P. Zhang, and H. Lu, “Mam- bapro: Multi-modal object re-identification with mamba aggregation and synergistic prompt,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 8, 2025, pp. 8150–8158

    2025

  26. [34]

    Logical relation inference and multiview information interaction for domain adaptation person re-identification,

    S. Li, F. Li, J. Li, H. Li, B. Zhang, D. Tao, and X. Gao, “Logical relation inference and multiview information interaction for domain adaptation person re-identification,”IEEE Transactions on Neural Networks and Learning Systems, 2023

    2023

  27. [35]

    Attribute-aligned domain- invariant feature learning for unsupervised domain adaptation person re-identification,

    H. Li, Y . Chen, D. Tao, Z. Yu, and G. Qi, “Attribute-aligned domain- invariant feature learning for unsupervised domain adaptation person re-identification,”IEEE Transactions on Information Forensics and Security, vol. 16, pp. 1480–1494, 2020

    2020

  28. [36]

    Hybrid contrastive learning for un- supervised person re-identification,

    T. Si, F. He, Z. Zhang, and Y . Duan, “Hybrid contrastive learning for un- supervised person re-identification,”IEEE Transactions on Multimedia, vol. 25, pp. 4323–4334, 2022. IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 15

    2022

  29. [37]

    Learn- ing to reduce dual-level discrepancy for infrared-visible person re- identification,

    Z. Wang, Z. Wang, Y . Zheng, Y .-Y . Chuang, and S. Satoh, “Learn- ing to reduce dual-level discrepancy for infrared-visible person re- identification,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2019, pp. 618–626

    2019

  30. [38]

    Tri-modality consistency optimization with heterogeneous augmented images for visible-infrared person re- identification,

    T. Si, F. He, P. Li, and X. Gao, “Tri-modality consistency optimization with heterogeneous augmented images for visible-infrared person re- identification,”Neurocomputing, vol. 523, pp. 170–181, 2023

    2023

  31. [39]

    Infrared-visible cross-modal person re-identification with an x modality,

    D. Li, X. Wei, X. Hong, and Y . Gong, “Infrared-visible cross-modal person re-identification with an x modality,” inProceedings of the AAAI conference on Artificial Intelligence, vol. 34, no. 04, 2020, pp. 4610– 4617

    2020

  32. [40]

    Hi-cmd: Hierar- chical cross-modality disentanglement for visible-infrared person re- identification,

    S. Choi, S. Lee, Y . Kim, T. Kim, and C. Kim, “Hi-cmd: Hierar- chical cross-modality disentanglement for visible-infrared person re- identification,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2020, pp. 10 257–10 266

    2020

  33. [41]

    Syncretic modality collabora- tive learning for visible infrared person re-identification,

    Z. Wei, X. Yang, N. Wang, and X. Gao, “Syncretic modality collabora- tive learning for visible infrared person re-identification,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 225–234

    2021

  34. [42]

    Discover cross-modality nuances for visible-infrared person re- identification,

    Q. Wu, P. Dai, J. Chen, C.-W. Lin, Y . Wu, F. Huang, B. Zhong, and R. Ji, “Discover cross-modality nuances for visible-infrared person re- identification,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2021, pp. 4330–4339

    2021

  35. [43]

    Learning memory-augmented unidirectional metrics for cross-modality person re- identification,

    J. Liu, Y . Sun, F. Zhu, H. Pei, Y . Yang, and W. Li, “Learning memory-augmented unidirectional metrics for cross-modality person re- identification,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2022, pp. 19 366–19 375

    2022

  36. [44]

    Mrcn: A novel modality restitution and compensation network for visible-infrared person re- identification,

    Y . Zhang, Y . Yan, J. Li, and H. Wang, “Mrcn: A novel modality restitution and compensation network for visible-infrared person re- identification,”arXiv preprint arXiv:2303.14626, 2023

    2023 arXiv

  37. [45]

    Diverse embedding expansion network and low-light cross-modality benchmark for visible-infrared person re- identification,

    Y . Zhang and H. Wang, “Diverse embedding expansion network and low-light cross-modality benchmark for visible-infrared person re- identification,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2023, pp. 2153–2162

    2023

  38. [46]

    Cross- modality person re-identification with memory-based contrastive embed- ding,

    D. Cheng, X. Wang, N. Wang, Z. Wang, X. Wang, and X. Gao, “Cross- modality person re-identification with memory-based contrastive embed- ding,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 1, 2023, pp. 425–432

    2023

  39. [47]

    Diversity feature constraint based on heterogeneous data for unsupervised person re-identification,

    T. Si, F. He, P. Li, Y . Song, and L. Fan, “Diversity feature constraint based on heterogeneous data for unsupervised person re-identification,” Information Processing & Management, vol. 60, no. 3, p. 103304, 2023

    2023

  40. [48]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational Conference on Machine Learning. PMLR, 2021, pp. 8748–8763

    2021

  41. [49]

    Zegclip: Towards adapting clip for zero-shot semantic segmentation,

    Z. Zhou, Y . Lei, B. Zhang, L. Liu, and Y . Liu, “Zegclip: Towards adapting clip for zero-shot semantic segmentation,” inIEEE Conference on Computer Vision and Pattern Recognition, CVPR, 2023, pp. 11 175– 11 185

    2023

  42. [50]

    X-clip: End-to- end multi-grained contrastive learning for video-text retrieval,

    Y . Ma, G. Xu, X. Sun, M. Yan, J. Zhang, and R. Ji, “X-clip: End-to- end multi-grained contrastive learning for video-text retrieval,” inACM International Conference on Multimedia, 2022, pp. 638–647

    2022

  43. [51]

    Clip4caption: Clip for video caption,

    M. Tang, Z. Wang, Z. Liu, F. Rao, D. Li, and X. Li, “Clip4caption: Clip for video caption,” inACM International Conference on Multimedia, 2021, pp. 4858–4862

    2021

  44. [52]

    Towards open-vocabulary remote sensing image semantic segmentation,

    C. Ye, Y . Zhuge, and P. Zhang, “Towards open-vocabulary remote sensing image semantic segmentation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 9, 2025, pp. 9436– 9444

    2025

  45. [53]

    Multi-modal understanding and generation for object tracking,

    H. Zhu, P. Zhang, L. Xue, and G. Yuan, “Multi-modal understanding and generation for object tracking,”IEEE Transactions on Circuits and Systems for Video Technology, 2024

    2024

  46. [54]

    Vision-language tracking with clip and interactive prompt learning,

    H. Zhu, Q. Lu, L. Xue, P. Zhang, and G. Yuan, “Vision-language tracking with clip and interactive prompt learning,”IEEE Transactions on Intelligent Transportation Systems, 2024

    2024

  47. [55]

    Clip-driven fine-grained text- image person re-identification,

    S. Yan, N. Dong, L. Zhang, and J. Tang, “Clip-driven fine-grained text- image person re-identification,”IEEE Transactions on Image Processing, 2023

    2023

  48. [56]

    Cross-modal implicit relation reasoning and aligning for text-to-image person retrieval,

    D. Jiang and M. Ye, “Cross-modal implicit relation reasoning and aligning for text-to-image person retrieval,” inIEEE Conference on Computer Vision and Pattern Recognition, 2023, pp. 2787–2797

    2023

  49. [57]

    Region generation and assessment network for occluded person re- identification,

    S. He, W. Chen, K. Wang, H. Luo, F. Wang, W. Jiang, and H. Ding, “Region generation and assessment network for occluded person re- identification,”IEEE Transactions on Information Forensics and Secu- rity, 2023

    2023

  50. [58]

    Climb- reid: A hybrid clip-mamba framework for person re-identification,

    C. Yu, X. Liu, J. Zhu, Y . Wang, P. Zhang, and H. Lu, “Climb- reid: A hybrid clip-mamba framework for person re-identification,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 9, 2025, pp. 9589–9597

    2025

  51. [59]

    Clip-driven semantic discovery network for visible-infrared person re-identification,

    X. Yu, N. Dong, L. Zhu, H. Peng, and D. Tao, “Clip-driven semantic discovery network for visible-infrared person re-identification,”arXiv preprint arXiv:2401.05806, 2024

    2024 arXiv

  52. [60]

    Image-specific information suppression and implicit local alignment for text-based person search,

    S. Yan, H. Tang, L. Zhang, and J. Tang, “Image-specific information suppression and implicit local alignment for text-based person search,” IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 12, pp. 17 973–17 986, 2024

    2024

  53. [61]

    Breaking the paired sample barrier in person re-identification: Leveraging unpaired samples for domain generalization,

    H. Li, Y . Liu, Y . Zhang, J. Li, and Z. Yu, “Breaking the paired sample barrier in person re-identification: Leveraging unpaired samples for domain generalization,”IEEE Transactions on Information Forensics and Security, 2025

    2025

  54. [62]

    Learning by aligning: Visible- infrared person re-identification using cross-modal correspondences,

    H. Park, S. Lee, J. Lee, and B. Ham, “Learning by aligning: Visible- infrared person re-identification using cross-modal correspondences,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 12 046–12 055

    2021

  55. [63]

    Farewell to mutual information: Variational distillation for cross-modal person re- identification,

    X. Tian, Z. Zhang, S. Lin, Y . Qu, Y . Xie, and L. Ma, “Farewell to mutual information: Variational distillation for cross-modal person re- identification,” inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2021, pp. 1522–1531

    2021

  56. [64]

    Channel augmented joint learning for visible-infrared recognition,

    M. Ye, W. Ruan, B. Du, and M. Z. Shou, “Channel augmented joint learning for visible-infrared recognition,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 13 567–13 576

    2021

  57. [65]

    Shape-erased feature learning for visible-infrared person re-identification,

    J. Feng, A. Wu, and W.-S. Zheng, “Shape-erased feature learning for visible-infrared person re-identification,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 752–22 761

    2023

  58. [66]

    Rgb-infrared cross-modality person re-identification via joint pixel and feature align- ment,

    G. Wang, T. Zhang, J. Cheng, S. Liu, Y . Yang, and Z. Hou, “Rgb-infrared cross-modality person re-identification via joint pixel and feature align- ment,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 3623–3632

    2019

  59. [67]

    Dynamic dual-attentive aggregation learning for visible-infrared person re- identification,

    M. Ye, J. Shen, D. J. Crandall, L. Shao, and J. Luo, “Dynamic dual-attentive aggregation learning for visible-infrared person re- identification,” inComputer Vision–ECCV 2020: 16th European Con- ference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVII 16. Springer, 202...

    2020

  60. [68]

    Towards a unified middle modal- ity learning for visible-infrared person re-identification,

    Y . Zhang, Y . Yan, Y . Lu, and H. Wang, “Towards a unified middle modal- ity learning for visible-infrared person re-identification,” inProceedings of the 29th ACM international Conference on Multimedia, 2021, pp. 788–796

    2021

  61. [69]

    Learning with twin noisy labels for visible-infrared person re-identification,

    M. Yang, Z. Huang, P. Hu, T. Li, J. Lv, and X. Peng, “Learning with twin noisy labels for visible-infrared person re-identification,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 14 308–14 317

    2022

  62. [70]

    Tran- sreid: Transformer-based object re-identification,

    S. He, H. Luo, P. Wang, F. Wang, H. Li, and W. Jiang, “Tran- sreid: Transformer-based object re-identification,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 15 013–15 022

    2021

  63. [71]

    Learning deep features for discriminative localization,

    B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, and A. Torralba, “Learning deep features for discriminative localization,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 2921–2929

    2016

Pith tools

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