pith. sign in

arxiv: 2512.03745 · v2 · submitted 2025-12-03 · 💻 cs.CV

Dual-level Modality Debiasing Learning for Unsupervised Visible-Infrared Person Re-Identification

Jiaze Li , Yan Lu , Bin Liu , Guojun Yin , Mang Ye This is my paper

Pith reviewed 2026-05-17 02:24 UTC · model grok-4.3

classification 💻 cs.CV
keywords unsupervised visible-infrared person re-identificationmodality debiasingcausal modelingmodality-invariant featurestwo-stage learning pipelinefeature alignment
0
0 comments X

The pith

Dual-level debiasing at model and optimization stages removes modality bias in unsupervised visible-infrared person re-identification.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper targets modality bias that arises when single-modality training precedes cross-modality learning in unsupervised visible-infrared person re-identification pipelines. Single-modality cues naturally carry forward and degrade identity discrimination. The proposed Dual-level Modality Debiasing Learning framework counters this by intervening at the model level with a causality-inspired adjustment module and at the optimization level with a collaborative training strategy that combines augmentation, label refinement, and feature alignment. If the approach works, the resulting model learns features that ignore modality-specific cues and generalize better across visible and infrared spectra on standard benchmarks.

Core claim

The authors establish that a two-level intervention—replacing likelihood-based modeling with causal modeling inside the Causality-inspired Adjustment Intervention module and applying Collaborative Bias-free Training across data, labels, and features—interrupts modality bias propagation, produces low-biased representations, and yields modality-invariant features together with a more generalized model.

What carries the argument

The Dual-level Modality Debiasing Learning framework, whose load-bearing components are the Causality-inspired Adjustment Intervention module that substitutes causal modeling for likelihood modeling and the Collaborative Bias-free Training strategy that coordinates modality-specific augmentation with label refinement and feature alignment.

If this is right

  • Modality-specific cues learned during single-modality training no longer propagate into cross-modality stages.
  • Identity discrimination improves because representations become less contaminated by modality artifacts.
  • The resulting model shows stronger generalization across visible and infrared images on existing benchmark datasets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same two-level structure could be tested on other unsupervised multi-modal retrieval tasks such as visible-thermal or RGB-depth matching.
  • Integrating the causal intervention with existing pseudo-label refinement methods might further stabilize training without extra supervision.
  • Running the framework on datasets with extreme lighting variation would reveal whether the learned invariance holds under realistic surveillance conditions.

Load-bearing premise

Replacing likelihood-based modeling with causal modeling inside the adjustment module actually stops modality-induced spurious patterns from entering the learned representations.

What would settle it

A test that extracts and compares modality-specific cues from features produced by the trained model versus a standard baseline would falsify the claim if the cues remain equally strong after the dual-level debiasing steps.

Figures

Figures reproduced from arXiv: 2512.03745 by Bin Liu, Guojun Yin, Jiaze Li, Mang Ye, Yan Lu.

Figure 1
Figure 1. Figure 1: Existing USL-VI-ReID methods suffer from modality bias, leading to modality-related features. In contrast, our approach achieves modality-invariant feature learning through causal modeling and unbiased optimization. Green, yellow, and blue circles represent visible-specific, infrared-specific, and modality￾shared information, respectively. To address the aforementioned modality bias issue, we propose a Dua… view at source ↗
Figure 2
Figure 2. Figure 2: The framework of the proposed DMDL. After obtaining cross-modality pseudo-labels through [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) The structural causal model in cross-modality learning for USL-VI-ReID. (b) The modified [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the modality-specific augmentation. Circles represent channels of images. Subscript [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Detailed analysis of CBT on the SYSU-MM01 dataset under (a) all-search and (b) indoor-search [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Parameter analysis of λcai and λf a on the SYSU-MM01 dataset (all-search). (a) Baseline (b) DMDL wrong right [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The t-SNE (first row) and similarity distribution (second row) visualization of 20 randomly selected [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Cross-modality pseudo-label quality analysis over di [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of the retrieval results obtained by the baseline and our DMDL on the SYSU-MM01 [PITH_FULL_IMAGE:figures/full_fig_p023_9.png] view at source ↗
read the original abstract

Two-stage learning pipeline has achieved promising results in unsupervised visible-infrared person re-identification (USL-VI-ReID). It first performs single-modality learning and then operates cross-modality learning to tackle the modality discrepancy. Although promising, this pipeline inevitably introduces modality bias: modality-specific cues learned in the single-modality training naturally propagate into the following cross-modality learning, impairing identity discrimination and generalization. To address this issue, we propose a Dual-level Modality Debiasing Learning (DMDL) framework that implements debiasing at both the model and optimization levels. At the model level, we propose a Causality-inspired Adjustment Intervention (CAI) module that replaces likelihood-based modeling with causal modeling, preventing modality-induced spurious patterns from being introduced, leading to a low-biased model. At the optimization level, a Collaborative Bias-free Training (CBT) strategy is introduced to interrupt the propagation of modality bias across data, labels, and features by integrating modality-specific augmentation, label refinement, and feature alignment. Extensive experiments on benchmark datasets demonstrate that DMDL could enable modality-invariant feature learning and a more generalized model. The code is available at https://github.com/priester3/DMDL.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript claims that two-stage pipelines for unsupervised visible-infrared person re-identification (USL-VI-ReID) introduce modality bias that propagates from single-modality pretraining into cross-modality learning, impairing identity discrimination. It proposes the Dual-level Modality Debiasing Learning (DMDL) framework to address this via a Causality-inspired Adjustment Intervention (CAI) module at the model level, which replaces likelihood-based modeling with causal modeling to avoid introducing modality-induced spurious patterns, and a Collaborative Bias-free Training (CBT) strategy at the optimization level that combines modality-specific augmentation, label refinement, and feature alignment to interrupt bias propagation across data, labels, and features. Extensive experiments on benchmark datasets are reported to show that DMDL enables modality-invariant feature learning and improved generalization; code is released.

Significance. If the causal adjustment mechanism in CAI demonstrably removes modality-specific spurious correlations (rather than acting as generic regularization), the dual-level debiasing approach would represent a meaningful advance for cross-modal ReID by directly targeting bias propagation in two-stage pipelines. The combination of model-level causal intervention and optimization-level collaborative training, together with public code, would support reproducibility and could influence subsequent work on modality-invariant representations.

major comments (1)
  1. [CAI module] CAI module description: the central claim that replacing likelihood-based modeling with causal modeling prevents modality-induced spurious patterns from being introduced into representations requires an explicit causal graph, a defined intervention operator (e.g., do-calculus adjustment on the modality variable as confounder), or a proof that the adjustment targets modality-specific cues rather than performing standard feature reweighting or attention. Without this, it is unclear whether CAI achieves the stated causal debiasing; this is load-bearing because the CBT stage operates on CAI outputs and any residual modality cue would propagate into label refinement and alignment.
minor comments (2)
  1. [Abstract] The abstract states that extensive experiments support the claims yet provides no quantitative results, ablation details, or error analysis; adding at least the top-line mAP/Rank-1 numbers on the primary benchmarks would allow readers to assess the magnitude of improvement immediately.
  2. [Method sections] Ensure that all equations and algorithmic steps in the CAI and CBT sections use consistent notation and explicitly define any new symbols or operators introduced for the causal adjustment and bias-free training components.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The major comment raises an important point about clarifying the causal mechanism in the CAI module, which we address below. We believe incorporating the requested details will improve the rigor of the presentation.

read point-by-point responses

  1. Referee: [CAI module] CAI module description: the central claim that replacing likelihood-based modeling with causal modeling prevents modality-induced spurious patterns from being introduced into representations requires an explicit causal graph, a defined intervention operator (e.g., do-calculus adjustment on the modality variable as confounder), or a proof that the adjustment targets modality-specific cues rather than performing standard feature reweighting or attention. Without this, it is unclear whether CAI achieves the stated causal debiasing; this is load-bearing because the CBT stage operates on CAI outputs and any residual modality cue would propagate into label refinement and alignment.

    Authors: We agree that an explicit causal graph and formal intervention details would strengthen the exposition of the CAI module. In the revised manuscript we will add a dedicated figure depicting the causal graph in which modality serves as a confounder between the observed features and the identity label. We will also provide the mathematical formulation of the adjustment operator using do-calculus to intervene on the modality variable, together with a short derivation showing that the resulting representation removes modality-specific spurious correlations while preserving identity-discriminative information. This formulation distinguishes the operation from generic attention or reweighting by explicitly blocking the back-door path from modality to the prediction. Because the CBT stage is applied to the outputs of this adjusted representation, the added details will also clarify why residual modality cues are not expected to propagate into label refinement and feature alignment. revision: yes

Circularity Check

0 steps flagged

No significant circularity in DMDL derivation chain

full rationale

The paper introduces a two-stage pipeline critique and proposes DMDL with distinct CAI (causality-inspired adjustment) and CBT (collaborative bias-free training) components at model and optimization levels. No self-definitional constructs appear where outputs are defined in terms of inputs by construction. No fitted parameters from data subsets are relabeled as predictions. The central claims rest on novel module designs rather than load-bearing self-citations or imported uniqueness theorems. The abstract and description present the replacement of likelihood modeling with causal modeling and the integration of augmentation/refinement/alignment as independent contributions without reduction to prior fitted quantities or ansatz smuggling. The framework remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard unsupervised learning assumptions plus the domain-specific premise that modality bias propagates through two-stage pipelines; no new physical entities or free parameters are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Modality-specific cues learned in single-modality training propagate into cross-modality learning and impair identity discrimination.
    Explicitly stated in the abstract as the core problem the framework addresses.

pith-pipeline@v0.9.0 · 5523 in / 1082 out tokens · 55765 ms · 2026-05-17T02:24:29.823101+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages · 1 internal anchor

  1. [1]

    M. Ye, W. Ruan, B. Du, M. Z. Shou, Channel augmented joint learning for visible- infrared recognition, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 13567–13576

    work page 2021

  2. [2]

    K. Ren, L. Zhang, Implicit discriminative knowledge learning for visible-infrared person re-identification, in: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2024, pp. 393–402. 29

    work page 2024

  3. [3]

    B. Yang, M. Ye, J. Chen, Z. Wu, Augmented dual-contrastive aggregation learn- ing for unsupervised visible-infrared person re-identification, in: Proceedings of the 30th ACM International Conference on Multimedia, 2022, pp. 2843–2851

    work page 2022

  4. [4]

    Z. Wu, M. Ye, Unsupervised visible-infrared person re-identification via progres- sive graph matching and alternate learning, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 9548–9558

    work page 2023

  5. [5]

    Cheng, L

    D. Cheng, L. He, N. Wang, S. Zhang, Z. Wang, X. Gao, Efficient bilateral cross- modality cluster matching for unsupervised visible-infrared person reid, in: Pro- ceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 1325–1333

    work page 2023

  6. [6]

    J. Shi, X. Yin, Y . Zhang, Y . Xie, Y . Qu, et al., Learning commonality, divergence and variety for unsupervised visible-infrared person re-identification, Advances in Neural Information Processing Systems 37 (2024) 99715–99734

    work page 2024

  7. [7]

    X. Teng, L. Lan, D. Chen, K. Xu, N. Yin, Relieving universal label noise for un- supervised visible-infrared person re-identification by inferring from neighbors, in: Proceedings of the AAAI Conference on Artificial Intelligence, V ol. 39, 2025, pp. 7356–7364

    work page 2025

  8. [8]

    Z. Dai, G. Wang, W. Yuan, S. Zhu, P. Tan, Cluster contrast for unsupervised person re-identification, in: Proceedings of the Asian Conference on Computer Vision, 2022, pp. 1142–1160

    work page 2022

  9. [9]

    Cheng, X

    D. Cheng, X. Huang, N. Wang, L. He, Z. Li, X. Gao, Unsupervised visible- infrared person reid by collaborative learning with neighbor-guided label refine- ment, in: Proceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 7085–7093

    work page 2023

  10. [10]

    L. He, D. Cheng, N. Wang, X. Gao, Exploring homogeneous and heterogeneous consistent label associations for unsupervised visible-infrared person reid, Inter- national Journal of Computer Vision (2024) 1–20. 30

    work page 2024

  11. [11]

    M. Ye, Z. Wu, B. Du, Dual-level matching with outlier filtering for unsupervised visible-infrared person re-identification, IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    work page 2025

  12. [12]

    B. Yang, J. Chen, M. Ye, Towards grand unified representation learning for unsupervised visible-infrared person re-identification, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 11069– 11079

    work page 2023

  13. [13]

    Z. Pang, C. Wang, L. Zhao, Y . Liu, G. Sharma, Cross-modality hierarchical clus- tering and refinement for unsupervised visible-infrared person re-identification, IEEE Transactions on Circuits and Systems for Video Technology (2023)

    work page 2023

  14. [14]

    B. Yang, J. Chen, M. Ye, Shallow-deep collaborative learning for unsupervised visible-infrared person re-identification, in: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2024, pp. 16870–16879

    work page 2024

  15. [15]

    Pearl, M

    J. Pearl, M. Glymour, N. P. Jewell, Causal inference in statistics: A primer, John Wiley & Sons, 2016

    work page 2016

  16. [16]

    T. Kim, S. Shin, Y . Yu, H. G. Kim, Y . M. Ro, Causal mode multiplexer: A novel framework for unbiased multispectral pedestrian detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 26784–26793

    work page 2024

  17. [17]

    X. Li, Y . Lu, B. Liu, Y . Liu, G. Yin, Q. Chu, J. Huang, F. Zhu, R. Zhao, N. Yu, Counterfactual intervention feature transfer for visible-infrared person re- identification, in: European Conference on Computer Vision, Springer, 2022, pp. 381–398

    work page 2022

  18. [18]

    Zhang, Z

    Y .-F. Zhang, Z. Zhang, D. Li, Z. Jia, L. Wang, T. Tan, Learning domain invariant representations for generalizable person re-identification, IEEE Transactions on Image Processing 32 (2022) 509–523

    work page 2022

  19. [19]

    Z. Yang, M. Lin, X. Zhong, Y . Wu, Z. Wang, Good is bad: Causality inspired cloth-debiasing for cloth-changing person re-identification, in: Proceedings of 31 the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1472–1481

    work page 2023

  20. [20]

    X. Li, Y . Lu, B. Liu, Y . Hou, Y . Liu, Q. Chu, W. Ouyang, N. Yu, Clothes- invariant feature learning by causal intervention for clothes-changing person re- identification, arXiv preprint arXiv:2305.06145 (2023)

    work page arXiv 2023

  21. [21]

    X.-C. Li, X. Xia, F. Zhu, T. Liu, X.-Y . Zhang, C.-L. Liu, Dynamics-aware loss for learning with label noise, Pattern Recognition 144 (2023) 109835

    work page 2023

  22. [22]

    J. Han, P. Luo, X. Wang, Deep self-learning from noisy labels, in: Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 5138– 5147

    work page 2019

  23. [23]

    Huang, J

    Z. Huang, J. Zhang, H. Shan, Twin contrastive learning with noisy labels, in: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, 2023, pp. 11661–11670

    work page 2023

  24. [24]

    Zhang, Y

    X. Zhang, Y . Ge, Y . Qiao, H. Li, Refining pseudo labels with clustering consen- sus over generations for unsupervised object re-identification, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 3436–3445

    work page 2021

  25. [25]

    Q. He, Z. Wang, Z. Zheng, H. Hu, Spatial and temporal dual-attention for unsu- pervised person re-identification, IEEE Transactions on Intelligent Transportation Systems 25 (2) (2023) 1953–1965

    work page 2023

  26. [26]

    Y . Cho, W. J. Kim, S. Hong, S.-E. Yoon, Part-based pseudo label refinement for unsupervised person re-identification, in: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, 2022, pp. 7308–7318

    work page 2022

  27. [27]

    J. Shi, Y . Zhang, X. Yin, Y . Xie, Z. Zhang, J. Fan, Z. Shi, Y . Qu, Dual pseudo- labels interactive self-training for semi-supervised visible-infrared person re- identification, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 11218–11228. 32

    work page 2023

  28. [28]

    J. Shi, X. Yin, Y . Chen, Y . Zhang, Z. Zhang, Y . Xie, Y . Qu, Multi-memory match- ing for unsupervised visible-infrared person re-identification, in: European Con- ference on Computer Vision, Springer, 2024, pp. 456–474

    work page 2024

  29. [29]

    In Defense of the Triplet Loss for Person Re-Identification

    A. Hermans, L. Beyer, B. Leibe, In defense of the triplet loss for person re- identification, arXiv preprint arXiv:1703.07737 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  30. [30]

    Arazo, D

    E. Arazo, D. Ortego, P. Albert, N. O’Connor, K. McGuinness, Unsupervised la- bel noise modeling and loss correction, in: International conference on machine learning, PMLR, 2019, pp. 312–321

    work page 2019

  31. [31]

    K. Xu, J. Ba, R. Kiros, K. Cho, A. Courville, R. Salakhudinov, R. Zemel, Y . Ben- gio, Show, attend and tell: Neural image caption generation with visual attention, in: International conference on machine learning, PMLR, 2015, pp. 2048–2057

    work page 2015

  32. [32]

    Jambigi, R

    C. Jambigi, R. Rawal, A. Chakraborty, Mmd-reid: A simple but effective solution for visible-thermal person reid, arXiv preprint arXiv:2111.05059 (2021)

    work page arXiv 2021

  33. [33]

    Wu, W.-S

    A. Wu, W.-S. Zheng, H.-X. Yu, S. Gong, J. Lai, Rgb-infrared cross-modality person re-identification, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 5380–5389

    work page 2017

  34. [34]

    D. T. Nguyen, H. G. Hong, K. W. Kim, K. R. Park, Person recognition system based on a combination of body images from visible light and thermal cameras, Sensors 17 (3) (2017) 605

    work page 2017

  35. [35]

    Zhang, H

    Y . Zhang, H. Wang, Diverse embedding expansion network and low-light cross- modality benchmark for visible-infrared person re-identification, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 2153–2162

    work page 2023

  36. [36]

    M. Ye, J. Shen, G. Lin, T. Xiang, L. Shao, S. C. Hoi, Deep learning for person re-identification: A survey and outlook, IEEE transactions on pattern analysis and machine intelligence 44 (6) (2021) 2872–2893. 33

    work page 2021

  37. [37]

    X. Wang, R. Girshick, A. Gupta, K. He, Non-local neural networks, in: Proceed- ings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7794–7803

    work page 2018

  38. [38]

    Radenovi ´c, G

    F. Radenovi ´c, G. Tolias, O. Chum, Fine-tuning cnn image retrieval with no human annotation, IEEE transactions on pattern analysis and machine intelligence 41 (7) (2018) 1655–1668

    work page 2018

  39. [39]

    C. Chen, M. Ye, M. Qi, J. Wu, J. Jiang, C.-W. Lin, Structure-aware positional transformer for visible-infrared person re-identification, IEEE Transactions on Image Processing 31 (2022) 2352–2364

    work page 2022

  40. [40]

    Zhang, C

    Q. Zhang, C. Lai, J. Liu, N. Huang, J. Han, Fmcnet: Feature-level modality compensation for visible-infrared person re-identification, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7349–7358

    work page 2022

  41. [41]

    H. Yu, X. Cheng, W. Peng, W. Liu, G. Zhao, Modality unifying network for visible-infrared person re-identification, in: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, 2023, pp. 11185–11195

    work page 2023

  42. [42]

    J. Shi, X. Yin, D. Zhang, Z. Zhang, Y . Xie, Y . Qu, Two-stage knowledge dis- tillation for visible-infrared person re-identification, Pattern Recognition (2025) 111850

    work page 2025

  43. [43]

    J. Wang, Z. Zhang, M. Chen, Y . Zhang, C. Wang, B. Sheng, Y . Qu, Y . Xie, Op- timal transport for label-efficient visible-infrared person re-identification, in: Eu- ropean Conference on Computer Vision, Springer, 2022, pp. 93–109

    work page 2022

  44. [44]

    X. Zhu, L. Dong, X. Chen, X. Zhang, F. Qi, X.-Y . Jing, Confidence guided semi-supervised cross-modality person re-identification, Pattern Recognition 165 (2025) 111669

    work page 2025

  45. [45]

    Y . Yang, W. Hu, H. Hu, Progressive cross-modal association learning for unsuper- vised visible-infrared person re-identification, IEEE Transactions on Information Forensics and Security (2025). 34

    work page 2025

  46. [46]

    Y . Li, Y . Sun, Y . Qin, D. Peng, X. Peng, P. Hu, Robust dual- ity learning for unsupervised visible-infrared person re-identification, IEEE Transactions on Information Forensics and Security 20 (2025) 1937–1948. doi:10.1109/TIFS.2025.3536613

    work page doi:10.1109/tifs.2025.3536613 2025

  47. [47]

    Cheng, L

    D. Cheng, L. He, N. Wang, D. Zhang, X. Gao, Semantic-aligned learning with collaborative refinement for unsupervised vi-reid: D. cheng et al., International Journal of Computer Vision (2025) 1–23

    work page 2025

  48. [48]

    H. Park, S. Lee, J. Lee, B. Ham, Learning by aligning: Visible-infrared per- son re-identification using cross-modal correspondences, in: Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 12046–12055

    work page 2021

  49. [49]

    M. Yang, Z. Huang, P. Hu, T. Li, J. Lv, 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. 14308–14317

    work page 2022

  50. [50]

    Y . Feng, F. Chen, G. Sun, F. Wu, Y . Ji, T. Liu, S. Liu, X.-Y . Jing, J. Luo, Learning multi-granularity representation with transformer for visible-infrared person re- identification, Pattern Recognition 164 (2025) 111510

    work page 2025

  51. [51]

    Z. Pang, L. Zhao, Y . Liu, G. Sharma, C. Wang, Inter-modality similarity learning for unsupervised multi-modality person re-identification, IEEE Transactions on Circuits and Systems for Video Technology 34 (10) (2024) 10411–10423

    work page 2024

  52. [52]

    van der Maaten, G

    L. van der Maaten, G. Hinton, Visualizing data using t-sne. journal of machine learning research 9, Nov (2008) (2008)

    work page 2008

  53. [53]

    Cournapeau, G

    D. Cournapeau, G. members, scikit-learn, https://scikit-learn.org/stable/index.html(2007)

    work page 2007

  54. [54]

    Ester, H.-P

    M. Ester, H.-P. Kriegel, J. Sander, X. Xu, et al., A density-based algorithm for discovering clusters in large spatial databases with noise, in: kdd, V ol. 96, 1996, pp. 226–231. 35

    work page 1996