TCMM: Token Constraint and Multi-Scale Memory Bank of Contrastive Learning for Unsupervised Person Re-identification

Chen-Kuo Chiang; Hsin-Che Chien; Zheng-An Zhu

REVIEW 4 major objections 4 minor 50 references

TCMM: Token Constraint and Multi-Scale Memory Bank of Contrastive Learning for Unsupervised Person Re-identification

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that coupling a ViT token constraint with a two-level contrastive memory bank sets a new state of the art in unsupervised person re-identification.

desk verdict Strong reported numbers, but a load-bearing flaw in the memory update equations and test-set hyperparameter tuning make the SOTA claims unreliable as written. read the letter →

arxiv 2501.09044 v1 pith:T5GTFUSW submitted 2025-01-15 cs.CV cs.AI

Zheng-An Zhu , Hsin-Che Chien , Chen-Kuo Chiang This is my paper · ORCID

classification cs.CVcs.AI

keywords unsupervisedpersonre-identificationcontrastivelearningvisiontransformertokenconstraintmulti-scalememorybankpseudolabelsprototypelossoutliersamples

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 unsupervised person re-identification can be pushed well past current benchmarks by adding two ideas to a ViT-based contrastive pipeline. First, a token constraint treats the patch token most similar to the global CLS feature as a positive and the R least similar patch tokens as negatives, a small contrastive loss meant to curb the damage of background clutter and occlusion patches inside the transformer. Second, a two-level memory bank—prototype memory built from cluster centroids and instance memory built from all samples including outliers—supplies two contrastive losses, one for feature consistency and one for learning from hard samples. The combination reaches 90.5% mAP / 96.0% Rank-1 on Market1501 and 52.0% / 78.4% on MSMT17, which the authors report as surpassing the previous state of the art by 2%/1.1% and 6.8%/3.3% respectively. A reader should care because accurate unsupervised re-identification matters for surveillance and retrieval settings where labeled training data is unavailable.

What carries the argument

The machinery is a trio of contrastive objectives on top of a ViT encoder. (1) ViT Token Constraint: for each image, the CLS feature is taken as anchor, the patch token with highest cosine similarity as positive, and the $R = \lfloor I\alpha \rfloor$ patch tokens with lowest similarity as negatives, in a softmax loss (Eq. 2). This operationalizes 'patch noises' as the least-aligned tokens. (2) Prototype Memory and Prototype Contrast Loss: cluster centroids $p_c = \frac{1}{|O_c|} \sum_{f_j \in O_c} f_j$ are stored in a momentum-updated memory; $L_{\text{proto}}$ is InfoNCE over these centroids, giving stable cluster-level positives and negatives. (3) Instance Memory and Anchor Contrastive Loss: an instance memory holds all features with pseudo labels, including outliers, and the loss uses the least similar same-cluster instance as positive and the $\kappa$ most similar different-cluster instances, possibly outliers, as negatives. Momentum updates for both memories keep bank features close to the current encoder.

What would settle it

Take a set of person images with occluders or heavy background; use the trained encoder to record, for each image, the patch token with highest cosine similarity to the CLS feature, and check with a person segmentation mask or bounding box whether that token lies on the pedestrian. If a substantial fraction of selected positives fall on background or occluder regions, then the token constraint is actively pulling the global representation toward clutter, contrary to its design.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the TCMM objective—$L_{\text{total}} = \lambda_{\text{con}} L_{\text{constraint}} + \lambda_{\text{pro}} L_{\text{proto}} + \lambda_{\text{an}} L_{\text{anchor}}$—produces a feature space in which same-identity images cluster tightly and different identities separate, enough to set a new state of the art on Market1501 and MSMT17. The token constraint term $L_{\text{constraint}}$ compares the CLS feature $f_b$ against all patch-token features, pulling the most similar token closer and pushing the $R$ least similar tokens away. The prototype term $L_{\text{proto}}$ applies InfoNCE between $f_b$ and cluster centroids stored in prototype memory, while the anchor term $L_{\text{anchor}}$ replaces the positive with the hardest in-class instance in memory and the negatives with the $\kappa$ most confusable out-of-class instances, including outliers. Ablations on MSMT17 show each term contributes decisively: the baseline reaches 25.8% mAP, adding the token constraint gives 37.5%, adding the prototype loss gives 42.0%, and the full model reaches 52.0% mAP with 78.4% Rank-1.

Load-bearing premise

The method assumes that, in every image, the patch token whose feature is most similar to the global image feature is genuinely useful information, while the $R$ least similar patch tokens are noise to be suppressed; the paper provides no patch-level evidence that the selected positive is semantically meaningful rather than background clutter or an occluder.

Editorial extensions

If this is right

If the results hold, the new state of the art on both standard unsupervised re-ID benchmarks is set by a method that does not change the clustering or pseudo-label process, so the losses can be layered onto existing ViT pipelines.
Each of the three loss terms gives a large isolated gain on MSMT17—+11.7 mAP for the token constraint, +4.5 for the prototype loss, and +10.0 for the anchor loss—so the design is not carried by a single component.
Including outlier samples as candidates in the anchor loss improves mAP/Rank-1 by 1.7%/1.3% over discarding them, supporting the paper's claim that hard-to-cluster samples carry usable information.
Small negative-selection budgets work best: $\alpha = 0.075$ for token negatives and $\kappa = 4$ for anchor negatives beat larger values, consistent with the hypothesis that rare noise tokens and a few hard negatives are the right contrastive targets.
Because the method adds no extra network parameters and keeps pseudo-label generation intact, practitioners can drop the three losses into an existing ViT-based unsupervised re-ID training loop.

Reading between the lines

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

Editorial inference: the token constraint's positive selection—the patch most similar to the CLS feature—can be tested directly with patch-level supervision; on occluded images one could measure how often the selected token lies on the pedestrian rather than the occluder, and if it often lies off-body, the constraint may be aligning the model toward clutter.
Editorial inference: the feature-consistency argument for prototype memory predicts that TCMM's advantage over instance-only contrastive learning grows as batch size shrinks, since the prototype term is what stabilizes small-batch updates; an ablation varying batch size would test this.
Editorial inference: the same hard-positive and confusable-negative recipe could transfer to other unsupervised retrieval tasks with noisy clusters, such as vehicle re-identification or product retrieval, where outliers are common.
Editorial inference: the exact density-based clustering settings (eps and min_samples) are not listed in the paper; reporting them would let the community attribute the reported gains to the new losses versus clustering configuration.

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.

Desk Editor's Note

read the letter

Zhu et al. put together a sensible combination of known tricks for unsupervised person re-id: a ViT token-level contrastive constraint, a prototype memory with contrastive loss, and an instance memory with hard-positive/negative mining that keeps outliers in the loop. That specific configuration is not in the cited papers, and the ablation study is clearly laid out. The paper is easy to follow and does a fair job of positioning itself against recent work. If the numbers hold, the method is practically useful, especially on MSMT17.

The soft spots are real, though. The most serious is the momentum update in Eqs. (6) and (7). They say "for all" prototypes and "for all" instance-memory entries, each is pulled toward the current batch feature with μ=0.2. Literally, that collapses all memory entries to a single moving-average vector within a few iterations, which would make Lproto and Lanchor degenerate constants. The ablations attribute large gains to those terms, so the written algorithm cannot be what actually ran. The intended update was probably only for the matching prototype and the current batch's instances, but that is not what Eqs. (6)–(7) state, and no code is provided to disambiguate. That is a load-bearing specification error, not a cosmetic one.

Next, α and κ are selected directly on MSMT17, and the chosen values are then reported on MSMT17. There is no held-out validation or third dataset. No error bars or multiple seeds, either. The paper also misstates its own gains: Table 4 shows 1.4/0.9 on Market-1501 and 6.3/8.3 on MSMT17, while Section 4.3 claims 2/1.1 and 6.8/3.3. That kind of arithmetic mismatch makes the headline numbers harder to trust. DBSCAN eps and min_samples are not reported, so the pseudo-label quality is unverifiable. The token constraint's assumption that the most CLS-similar patch is a semantic positive is plausible but untested; a patch-level oracle or a few visualizations would help.

None of this disproves the method. The conceptual design is coherent, the citations look appropriate, and the cleanest interpretation is that the authors wrote the update equations sloppily rather than that the whole system is fraudulent. But as it stands, the paper does not support the SOTA claim. It deserves a serious referee who will ask for the code, corrected equations, multi-seed runs, and an honest validation protocol. If those come back clean, this is a solid incremental contribution worth publishing. I would not cite it in its current form.

Referee Report

4 major / 4 minor

Summary. The paper proposes TCMM, an unsupervised person re-identification method combining a ViT token constraint, a prototype-level memory with a prototype contrast loss, and an instance-level memory with an anchor contrastive loss. The method uses DBSCAN pseudo-labels and momentum-updated memory banks, and reports state-of-the-art results on Market-1501 (90.5% mAP / 96.0% Rank-1) and MSMT17 (52.0% / 78.4%). The central claim is that each of the three loss terms contributes substantial gains over the baseline, and that the full system surpasses prior unsupervised re-id methods.

Significance. If the results hold, TCMM would be a simple and effective combination of existing ideas (token-level contrastive constraints, prototype and instance memory banks) that pushes unsupervised person re-id performance notably forward on MSMT17. The paper explicitly avoids extra model components and retains the standard pseudo-label pipeline, which would make the gains practically attractive. However, the empirical claim currently rests on an under-specified and internally questionable algorithm specification, hyperparameters tuned directly on the test sets, and no measures of variance. The manuscript also provides no code release despite the stated project link, so the central claims are not independently verifiable from the text alone.

major comments (4)

[Section 3.2.3, Eqs. (6) and (7)] The momentum update equations as written are internally inconsistent with the reported results. Eq. (6) updates every prototype p_c in M_proto using every batch feature f_b, and Eq. (7) updates every instance feature f_n in M_ins using every batch feature f_b, with mu = 0.2. If implemented literally, all memory entries are dragged toward all batch features each iteration and collapse to a common vector, making Lproto in Eq. (4) and Lanchor in Eq. (5) degenerate and unable to produce the gains attributed to them in Table 3. The most plausible reading is that only the entries corresponding to the current batch (or the matched pseudo-class prototype and matched instance) are updated, but this is not what the equations state and no code is provided to disambiguate. This is a load-bearing reproducibility flaw; the paper must state the exact update rule and ideally release code.
[Section 4.2 and Table 4] The hyperparameters alpha (Eq. (1)) and kappa (Eq. (5)) are tuned directly on MSMT17 and the best settings are then used to report the final MSMT17 numbers. Since there is no held-out validation split and no multiple-seed results, the reported SOTA comparisons are optimistically biased. Additionally, the claimed gains in Section 4.3 do not match Table 4: on Market-1501, TCMM (90.5/96.0) versus the best previous method ACFL-VIT (89.1/95.1) is an improvement of 1.4/0.9, not 2.0/1.1; on MSMT17, comparing against HSP-MFL (45.2/75.1) gives 6.8/3.3, but comparing against the strongest prior ACFL-VIT (45.7/70.1) gives 6.3/8.3. The text should consistently state the actual margins against the best prior method on each benchmark and should report standard deviations across at least three runs.
[Section 3.1, Eq. (2)] The token constraint assumes that the patch token with the highest similarity to the CLS feature is a reliable positive and that the R least similar tokens are noise. This assumption is not tested; the paper provides no patch-level oracle, no visualization, and no ablation that would show the selected positive tokens are semantically meaningful rather than background clutter or occluders that correlate with the global feature. Given that Lconstraint contributes the largest single gain in Table 3 (+11.7 mAP), the validity of this selection rule needs direct evidence.
[Algorithm 1 and Section 4.1] The clustering hyperparameters for DBSCAN (eps and min_samples) are never reported, despite the method's dependence on pseudo-labels from DBSCAN for both memory banks. Without these values, the experiments cannot be reproduced and the sensitivity of the results to clustering quality is unknown. The authors should report the DBSCAN settings and ideally show results across a range of eps values.

minor comments (4)

[Throughout] The notation is sometimes inconsistent: f_b denotes batch instance features in Eq. (2) but the anchor in Eq. (5); the superscript/subscript style for instance features (f^M_n vs f_ins) is confusing. Please standardize the notation.
[Figure 6] The x-axis label and the description of the alpha sweep are unclear: the text says 'from 0.025 to 1' but the figure is not described in enough detail to see the sampling points; please clarify the axis and the variance across runs.
[Section 4.2, Table 2] The statement that 'Not using all outlier samples may lead to the model lacking a global understanding' is speculative; the table only shows a small performance gap. Consider tempering the causal claim.
[Section 2.3] The related work section lists many methods but does not clearly position TCMM relative to the closest memory-bank baselines (e.g., ClusterNCE, SPCL, RTMem) beyond stating general limitations. A more explicit comparison of the memory update and sample-selection mechanisms would help.

Circularity Check

1 steps flagged · score 6.0 of 10

MSMT17 SOTA result is the tuned hyperparameter value, not an independent prediction; Market1501 remains independent.

fitted input called prediction [Section 4.2 (alpha/kappa ablations) through Section 4.3 (SOTA comparison), Tables 1 and 4]
"The experiments in Figure 6 show that the best mAP performance is achieved when α is set to 0.075 with 52%, respectively. ... Whenκ=4, we achieve the best mAP/Rank-1 performance at 52.0%/78.4%. ... Our TCMM architecture achieves 90.5%/96.0% mAP and Rank-1 on the Market1501 dataset. On the other hand, our TCMM architecture achieves 52.0%/78.4% mAP and Rank-1 on the MSMT17 dataset."

The MSMT17 numbers reported as the final result in Section 4.3 (52.0/78.4) are exactly the values obtained in Section 4.2 by sweeping alpha and kappa on MSMT17 and retaining the best mAP/Rank-1. The SOTA claim on MSMT17 is therefore the selected maximum of the tuning curve, not an out-of-sample prediction; the 'surpasses SOTA' comparison on MSMT17 is forced by the selection procedure. This matches the fitted-input-called-prediction pattern: hyperparameter values are fitted to the benchmark and then renamed as the predicted benchmark result. Market1501 is not affected by this specific circularity because the reported sweep was run on MSMT17 only, although the chosen hyperparameters are then applied there.

full rationale

The paper's central claim is an empirical SOTA comparison, not a first-principles derivation whose conclusion is presupposed by its inputs. The three loss terms are standard contrastive objectives built on DBSCAN pseudo-labels; using the model's own pseudo-labels for training is self-referential in the ordinary unsupervised-learning sense, but it is not a circularity because the losses are not defined as the quantities they are claimed to predict. No load-bearing self-citation chain appears: the authors' own prior work [22] is cited only as one related-work entry. The concrete circular step is the MSMT17 protocol: the final MSMT17 mAP/Rank-1 (52.0/78.4) is the value selected by sweeping alpha and kappa on MSMT17 in Section 4.2, and Section 4.3 then presents this same tuned value as the SOTA result; that part of the benchmark claim is fitted rather than predicted. Market1501, where no such sweep was reported, remains an independent check, so the paper is only partially circular. The separate concern raised by the skeptical review, that Eqs. (6)-(7) literally update every memory entry with every batch feature and would collapse the memory, is an internal-consistency and reproducibility defect rather than a circularity; it does not make the SOTA claim circular by definition, but it does undermine the believability of the reported gains.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method relies on several hand-tuned scalars (α, κ, λs) and on unreported clustering parameters. The token-constraint positive/negative selection rule and the hard-positive selection rule are domain assumptions about patch semantics and pseudo-label quality, not facts established in the paper. No new physical or external entities are introduced.

free parameters (4)

α (negative patch token rate) = 0.075
Grid searched on MSMT17 (Figure 6); controls R = floor(I*α) negative tokens in L_constraint; tuned on the same dataset where final results are reported.
κ (number of hard negative samples) = 4
Grid searched on MSMT17 (Table 1); controls number of cross-class hard negatives in L_anchor.
λ_con, λ_pro, λ_an = 1, 1, 1
Set 'by empirical' without ablation; balances the three losses.
DBSCAN eps and min_samples = not reported
Clustering is central to pseudo-label generation and memory bank construction, but eps/min_samples are never specified, so the method is underdetermined.

assumptions (5)

standard math InfoNCE contrastive loss improves representation quality when positives are semantically related and negatives are unrelated (Eq. 2, 4, 5).
The paper builds all losses on the InfoNCE formulation [42]; this is a standard assumption.
domain assumption The patch token with highest similarity to the CLS feature is a useful positive and the R least similar tokens are noise (Section 3.1).
This is the load-bearing assumption of L_constraint; it is not validated with any experiment.
domain assumption Cluster prototypes computed from pseudo-labels are stable and consistent enough to serve as positives/negatives (Eq. 3-4).
Prototype quality depends on clustering quality; DBSCAN's parameters are unreported.
domain assumption Hardest-in-class positive (least similar same-cluster sample) carries a valid training signal despite possible label noise (Eq. 5).
The hardest positive may be a mis-clustered sample from another identity, which could reinforce errors.
standard math Momentum update with μ=0.2 keeps memory features consistent (Eq. 6-7).
Common technique from MoCo [44]; assumed to work as in prior work.

how reviews work

0 comments

Cite this review

Pith. "Pith review of TCMM: Token Constraint and Multi-Scale Memory Bank of Contrastive Learning for Unsupervised Person Re-identification." pith.science (2026). https://pith.science/paper/T5GTFUSW

@misc{pith2026250109044,
  author       = {Pith},
  title        = {Pith review of: TCMM: Token Constraint and Multi-Scale Memory Bank of Contrastive Learning for Unsupervised Person Re-identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T5GTFUSW}},
  note         = {Machine review of arXiv:2501.09044}
}

read the original abstract

This paper proposes the ViT Token Constraint and Multi-scale Memory bank (TCMM) method to address the patch noises and feature inconsistency in unsupervised person re-identification works. Many excellent methods use ViT features to obtain pseudo labels and clustering prototypes, then train the model with contrastive learning. However, ViT processes images by performing patch embedding, which inevitably introduces noise in patches and may compromise the performance of the re-identification model. On the other hand, previous memory bank based contrastive methods may lead data inconsistency due to the limitation of batch size. Furthermore, existing pseudo label methods often discard outlier samples that are difficult to cluster. It sacrifices the potential value of outlier samples, leading to limited model diversity and robustness. This paper introduces the ViT Token Constraint to mitigate the damage caused by patch noises to the ViT architecture. The proposed Multi-scale Memory enhances the exploration of outlier samples and maintains feature consistency. Experimental results demonstrate that our system achieves state-of-the-art performance on common benchmarks. The project is available at \href{https://github.com/andy412510/TCMM}{https://github.com/andy412510/TCMM}.

Figures

Figures reproduced from arXiv: 2501.09044 by the authors.

**Figure 2.** The overall architecture of TCMM. A ViT encoder processes the input data to obtain token fea [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

**Figure 3.** The process of ViT token constraint.The output CLS feature [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: The process of prototype contrastive loss with DBSCAN example. The input data [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]

**Figure 5.** Figure 5: The process of anchor contrastive loss with DBSCAN example. The input data [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]

**Figure 6.** Figure 6: Study of α in Lconstraint on MSMT17 dataset. This experiment evaluates the impact of α on model performance with fixed settings Lproto and Lanchor. When α is 0.075, the best mAP performance of 52% is achieved. settings of κ and calculate mAP and Rank-1. Experiment resu…

**Figure 7.** Figure 7: The t-SNE visualization results on MSMT17 dataset. The same color represents samples classified [PITH_FULL_IMAGE:figures/full_fig_p027_7.png]

**Figure 8.** Figure 8: The attention maps visualization results of our TCMM. From left to right, the scenarios represent [PITH_FULL_IMAGE:figures/full_fig_p028_8.png]

**Figure 9.** Figure 9: The ranking list visualization results. The first row is the ranking list from the TCMM, and the [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 47 canonical work pages

[1]

H. Wang, J. Shen, Y . Liu, Y . Gao, E. Gavves, Nformer: Robust person re- identification with neighbor transformer, in: Proceedings of the IEEE /CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 7297– 7307

work page 2022
[2]

Z. Wang, F. Zhu, S. Tang, R. Zhao, L. He, J. Song, Feature erasing and di ffusion network for occluded person re-identification, in: Proceedings of the IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 4754–4763

work page 2022
[3]

J. Yin, S. Zhang, J. Xie, Z. Ma, J. Guo, Unsupervised person re-identification via simultaneous clustering and mask prediction, Pattern Recognition 126 (2022) 108568

work page 2022
[4]

X. Wang, Z. Sun, A. Chehri, G. Jeon, Y . Song, A novel attention-driven frame- work for unsupervised pedestrian re-identification with clustering optimization, Pattern Recognition 146 (2024) 110045

work page 2024
[5]

Zhang, D

X. Zhang, D. Li, Z. Wang, J. Wang, E. Ding, J. Q. Shi, Z. Zhang, J. Wang, Implicit sample extension for unsupervised person re-identification, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 7369–7378

work page 2022
[6]

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. 30

work page 2022
[7]

T. He, L. Shen, Y . Guo, G. Ding, Z. Guo, Secret: Self-consistent pseudo label refinement for unsupervised domain adaptive person re-identification, in: Pro- ceedings of the AAAI conference on artificial intelligence, V ol. 36, 2022, pp. 879–887

work page 2022
[8]

J. Zhao, J. Liao, J. Yuan, Hsp-mfl: A high-level semantic property driven multi- task feature learning network for unsupervised person re-id, Journal of Visual Communication and Image Representation 93 (2023) 103828

work page 2023

Show all 50 references

[9]

X. Han, X. Yu, G. Li, J. Zhao, G. Pan, Q. Ye, J. Jiao, Z. Han, Rethinking sampling strategies for unsupervised person re-identification, IEEE Transactions on Image Processing 32 (2023) 29–42

2023
[10]

H. Ji, L. Wang, S. Zhou, W. Tang, N. Zheng, G. Hua, Transfer easy to hard: Adversarial contrastive feature learning for unsupervised person re-identification, Pattern Recognition 145 (2024) 109973

2024
[11]

Z. Chen, Z. Fan, Y . Chen, Y . Zhu, Camera-aware cluster-instance joint on- line learning for unsupervised person re-identification, Pattern Recognition 151 (2024) 110359

2024
[12]

K. Zhu, H. Guo, T. Yan, Y . Zhu, J. Wang, M. Tang, Pass: Part-aware self- supervised pre-training for person re-identification, in: Computer Vision – ECCV 2022, Springer Nature Switzerland, 2022, pp. 198–214

2022
[13]

Y . Ge, F. Zhu, D. Chen, R. Zhao, et al., Self-paced contrastive learning with hybrid memory for domain adaptive object re-id, Advances in neural information processing systems 33 (2020) 11309–11321

2020
[14]

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 (ACCV), 2022, pp. 1142–1160

2022
[15]

Y . Wu, T. Huang, H. Yao, C. Zhang, Y . Shao, C. Han, C. Gao, N. Sang, Multi- centroid representation network for domain adaptive person re-id, in: Proceedings of the AAAI conference on artificial intelligence, V ol. 36, 2022, pp. 2750–2758. 31

2022
[16]

C. Zou, Z. Chen, Z. Cui, Y . Liu, C. Zhang, Discrepant and multi-instance prox- ies for unsupervised person re-identification, in: Proceedings of the IEEE /CVF International Conference on Computer Vision (ICCV), 2023, pp. 11058–11068

2023
[17]

Dosovitskiy, L

A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, An image is worth 16x16 words: Transformers for image recognition at scale, in: International Conference on Learning Repr...

2021
[18]

Y . Guo, D. Stutz, B. Schiele, Improving robustness of vision transformers by re- ducing sensitivity to patch corruptions, in: Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 4108–4118

2023
[19]

X. Lai, Z. Tian, L. Jiang, S. Liu, H. Zhao, L. Wang, J. Jia, Semi-supervised semantic segmentation with directional context-aware consistency, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 1205–1214

2021
[20]

Mittal, M

S. Mittal, M. Tatarchenko, T. Brox, Semi-supervised semantic segmentation with high- and low-level consistency, IEEE Transactions on Pattern Analysis and Ma- chine Intelligence 43 (4) (2021) 1369–1379

2021
[21]

D. Kwon, S. Kwak, Semi-supervised semantic segmentation with error localiza- tion network, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 9957–9967

2022
[22]

Zhu, C.-K

Z.-A. Zhu, C.-K. Chiang, Generative extension positive pairs and improving sample selection based on contrastive learning for unsupervised person re- identification, in: ICASSP 2024 - 2024 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP), 202...

2024
[23]

Zhang, J

J. Zhang, J. Yang, J. Yu, J. Fan, Semisupervised image classification by mutual learning of multiple self-supervised models, International Journal of Intelligent Systems 37 (5) (2022) 3117–3141. 32

2022
[24]

Z. Hao, H. Ge, J. Huang, Research on an unsupervised person re-identification based on image quality enhancement method, Engineering Applications of Arti- ficial Intelligence 123 (2023) 106392

2023
[25]

Y . Zhao, Q. Shu, X. Shi, Dual-level contrastive learning for unsupervised person re-identification, Image and Vision Computing 129 (2023) 104607

2023
[26]

Y . Zhao, Q. Shu, X. Shi, J. Zhan, Unsupervised person re-identification by dy- namic hybrid contrastive learning, Image and Vision Computing 137 (2023) 104786

2023
[27]

S. Xu, L. Luo, J. Hu, B. Yang, S. Hu, Semantic driven attention network with at- tribute learning for unsupervised person re-identification, Knowledge-Based Sys- tems 252 (2022) 109354

2022
[28]

Zhang, Z

W. Zhang, Z. Wei, L. Huang, K. Xie, Q. Qin, Adaptive attention-aware network for unsupervised person re-identification, Neurocomputing 411 (2020) 20–31

2020
[29]

H. Du, L. He, P. Liu, X. Hao, Inter-domain fusion and intra-domain style nor- malization network for unsupervised domain adaptive person re-identification, Digital Signal Processing 133 (2023) 103848

2023
[30]

S. Chen, L. Qiu, Z. Tian, Y . Yan, D.-H. Wang, S. Zhu, Mtnet: Mutual tri-training network for unsupervised domain adaptation on person re-identification, Journal of Visual Communication and Image Representation 90 (2023) 103749

2023
[31]

A. Shah, N. Srivastava, M. Khare, Effect of reconstruction losses in discriminative and generative learning based networks for the person re-identification, Procedia Computer Science 218 (2023) 1994–2006

2023
[32]

J. Yu, J. Peng, K. Li, H. Wang, Hybrid partial-constrained learning with orthog- onality regularization for unsupervised person re-identification, Engineering Ap- plications of Artificial Intelligence 123 (2023) 106200

2023
[33]

J. Yin, X. Zhang, Z. Ma, J. Guo, Y . Liu, A real-time memory updating strategy for unsupervised person re-identification, IEEE Transactions on Image Processing (2023). 33

2023
[34]

Grainger, T

R. Grainger, T. Paniagua, X. Song, N. Cuntoor, M. W. Lee, T. Wu, Paca-vit: Learning patch-to-cluster attention in vision transformers, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 18568–18578

2023
[35]

Yu, H.-T

Y .-C. Yu, H.-T. Lin, Semi-supervised domain adaptation with source label adap- tation, in: Proceedings of the IEEE /CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 24100–24109

2023
[36]

Y . Chen, X. Tan, B. Zhao, Z. Chen, R. Song, J. Liang, X. Lu, Boosting semi- supervised learning by exploiting all unlabeled data, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 7548–7557

2023
[37]

H. Xu, L. Liu, Q. Bian, Z. Yang, Semi-supervised semantic segmentation with prototype-based consistency regularization, Advances in neural information pro- cessing systems 35 (2022) 26007–26020

2022
[38]

J. Jin, S. Wang, Z. Dong, X. Liu, E. Zhu, Deep incomplete multi-view cluster- ing with cross-view partial sample and prototype alignment, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 11600–11609

2023
[39]

J. Li, P. Zhou, C. Xiong, S. C. Hoi, Prototypical contrastive learning of unsuper- vised representations, in: ICLR, 2021

2021
[40]

Nauta, J

M. Nauta, J. Schl ¨otterer, M. van Keulen, C. Seifert, Pip-net: Patch-based in- tuitive prototypes for interpretable image classification, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 2744–2753

2023
[41]

Z. Wu, Y . Xiong, S. X. Yu, D. Lin, Unsupervised feature learning via non- parametric instance discrimination, in: International Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 3733–3742. 34

2018
[42]

A. v. d. Oord, Y . Li, O. Vinyals, Representation learning with contrastive predic- tive coding, arXiv preprint arXiv:1807.03748 (2018)

2018 arXiv
[43]

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

1996
[44]

K. He, H. Fan, Y . Wu, S. Xie, R. Girshick, Momentum contrast for unsupervised visual representation learning, in: International Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 9729–9738

2020
[45]

Tarvainen, H

A. Tarvainen, H. Valpola, Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results, Advances in neural information processing systems 30 (2017)

2017
[46]

Zheng, L

L. Zheng, L. Shen, L. Tian, S. Wang, J. Wang, Q. Tian, Scalable person re- identification: A benchmark, in: IEEE International Conference on Computer Vision, 2015

2015
[47]

L. Wei, S. Zhang, W. Gao, Q. 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
[48]

Caron, H

M. Caron, H. Touvron, I. Misra, H. J ´egou, J. Mairal, P. Bojanowski, A. Joulin, Emerging properties in self-supervised vision transformers, in: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 9650–9660

2021
[49]

D. Fu, D. Chen, J. Bao, H. Yang, L. Yuan, L. Zhang, H. Li, D. Chen, Unsu- pervised pre-training for person re-identification, in: International Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 14750–14759

2021
[50]

H. Fan, L. Zheng, C. Yan, Y . Yang, Unsupervised person re-identification: Clus- tering and fine-tuning, ACM Transactions on Multimedia Computing, Communi- cations, and Applications (TOMM) 14 (4) (2018) 1–18. 35

2018

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

H. Wang, J. Shen, Y . Liu, Y . Gao, E. Gavves, Nformer: Robust person re- identification with neighbor transformer, in: Proceedings of the IEEE /CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 7297– 7307

work page 2022

[2] [2]

Z. Wang, F. Zhu, S. Tang, R. Zhao, L. He, J. Song, Feature erasing and di ffusion network for occluded person re-identification, in: Proceedings of the IEEE /CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 4754–4763

work page 2022

[3] [3]

J. Yin, S. Zhang, J. Xie, Z. Ma, J. Guo, Unsupervised person re-identification via simultaneous clustering and mask prediction, Pattern Recognition 126 (2022) 108568

work page 2022

[4] [4]

X. Wang, Z. Sun, A. Chehri, G. Jeon, Y . Song, A novel attention-driven frame- work for unsupervised pedestrian re-identification with clustering optimization, Pattern Recognition 146 (2024) 110045

work page 2024

[5] [5]

Zhang, D

X. Zhang, D. Li, Z. Wang, J. Wang, E. Ding, J. Q. Shi, Z. Zhang, J. Wang, Implicit sample extension for unsupervised person re-identification, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 7369–7378

work page 2022

[6] [6]

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. 30

work page 2022

[7] [7]

T. He, L. Shen, Y . Guo, G. Ding, Z. Guo, Secret: Self-consistent pseudo label refinement for unsupervised domain adaptive person re-identification, in: Pro- ceedings of the AAAI conference on artificial intelligence, V ol. 36, 2022, pp. 879–887

work page 2022

[8] [8]

J. Zhao, J. Liao, J. Yuan, Hsp-mfl: A high-level semantic property driven multi- task feature learning network for unsupervised person re-id, Journal of Visual Communication and Image Representation 93 (2023) 103828

work page 2023