REVIEW 3 major objections 4 minor 28 references
Unsupervised Domain-Adaptive Person Re-identification Based on Attributes
T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read Unsupervised domain-adaptive person re-identification can be built on transferred pedestrian attributes, and the paper reports Rank-1 accuracy of 43.0% when adapting from DukeMTMC-reID to Market-1501, exceeding previous attribute-based…
desk verdict Useful RAP-to-ReID transfer idea undercut by a comparison that quietly changes the source domain and a degenerate-looking adversarial loss. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is a modified adversarial discriminative domain adaptation setup: a source feature encoder pretrained on attribute labels, a target encoder initialized from it, and a discriminator trained to tell source from target feature vectors. The adaptation feeds both source and target images into the target encoder and uses a least-squares adversarial loss so the target mapping becomes invariant to both domains. An additional multi-label attribute classifier, applied to source images only, contributes $\alpha \mathcal{L}_{\text{attr}}$ with $\alpha = 0.1$, preventing the attribute recognition signal from collapsing while the discriminator aligns the distributions. This joint objective is what carries the transfer.
What would settle it
One concrete test would be to take the adapted encoder, run it on target images whose attribute annotations already exist (27 for Market-1501, 22 for DukeMTMC-reID), and check whether the projected attribute classifier has any accuracy. If it performs near chance on attributes like clothing color while the ReID matching still works, then the attribute-consistency story is wrong; if both fail, the transfer premise is unsupported.
Extended reading notes
Core claim
The core claim is that an unsupervised domain-adaptive framework can learn a domain-invariant mapping whose feature vectors encode semantic attributes, and that these features are sufficient for person matching in a new domain with no labels. The source encoder is first trained with multi-label sigmoid cross-entropy on 70 selected attributes from the RAP dataset. Adaptation feeds both source and target images into the target encoder, pits it against a least-squares discriminator, and simultaneously trains an attribute classifier on source samples to prevent the recognized attribute structure from being forgotten. After adaptation, simple Euclidean distance between feature vectors supports matching and ranking. The reported numbers show three-to-four-point Rank-1 gains from adaptation and margins over previous attribute-based unsupervised transfer methods of 4.3 and 12.8 percentage points in one direction and 5.0 and 7.5 in the other.
Load-bearing premise
The load-bearing premise is that pedestrian attributes look and mean the same thing across camera domains, so a mapping trained on attribute labels in one dataset still encodes attributes that separate identities in another dataset with no labels.
Editorial extensions
If this is right
- Attribute annotations from recognition datasets can supply supervision for ReID without identity labels, as demonstrated on two target ReID datasets.
- Because no target labels are used, the pipeline can in principle extend to any new camera network as long as a labeled attribute source dataset exists.
- The observed stability gain from the extra source-image classifier implies that preserving auxiliary supervision during adversarial adaptation matters, not just distribution matching.
- Simple Euclidean distance suffices for matching after adaptation, so no learned metric is required in the target domain.
- The directional asymmetry in results—features adapted from Duke generally transfer better to Market than the reverse—points to source-domain diversity as a factor in how well the adapted features generalize.
Reading between the lines
- The attribute-consistency premise could be quantified by measuring cross-dataset attribute classifier accuracy using the existing attribute annotations on the two ReID datasets; low transferred attribute accuracy would suggest the ReID gains come from domain alignment rather than attribute semantics.
- Because the method uses no identity information at all, a natural extension would combine it with pseudo-label self-training or clustering in the target domain, which this paper does not explore.
- The smaller adaptation gains on the harder Duke target suggest a testable scaling prediction: as the source-target visual gap grows, adversarial attribute transfer should degrade, which could be checked against even more distant domains.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes an unsupervised domain-adaptive person re-identification (ReID) framework that transfers attribute-related features from an attribute-recognition dataset (RAP) to unlabeled ReID targets (Market-1501 and DukeMTMC-reID). The method first trains a CNN for multi-label attribute recognition on RAP, then adapts it to the target ReID domain using an adversarial discriminative domain adaptation objective, with an additional attribute classifier on source samples to preserve attribute recognition during adaptation. Experiments report improved Rank-1/mAP after adaptation (Table 1: 28.2% to 32.1% Rank-1 on Market-1501, 15.6% to 18.7% on DukeMTMC-reID) and comparisons with TJ-AIDL and MMFA (Table 2) that show higher Rank-1 and mAP when one ReID dataset is used as source and the other as target. The central claim is that attribute consistency across domains enables transfer from RAP to ReID, and that the proposed adversarial adaptation with an attribute classifier outperforms existing attribute-based unsupervised domain-adaptive ReID methods.
Significance. If the results hold, the work addresses a practically important problem: leveraging richly attribute-annotated pedestrian datasets to relieve the shortage of attribute labels in ReID datasets. The idea of using RAP as a source domain for unsupervised ReID adaptation is reasonable, and the added attribute classifier to counter attribute-recognition degradation during adaptation is a sensible contribution. The paper reports standard metrics (CMC, mAP) on two large ReID benchmarks and gives clear ablation-style evidence in Table 1 and Fig. 4 that adaptation improves over no adaptation, with the additional classifier providing stability. However, the primary claimed advantage over state-of-the-art methods is not supported by the experimental protocol as presented, because Table 2 uses ReID datasets (Market/Duke) rather than RAP as the source, and Table 1 has no baseline comparison. In addition, the adversarial objective in Eq. (3) is degenerate as written and does not implement the stated domain-invariance goal. These are load-bearing issues that require either corrected experiments and derivations or a substantial reframing of the claims.
major comments (3)
- [Section 3, Eq. (3)] The first expectation is over Xa∪Xp with target label 1 for D(M(x)); the second is over Xa with target label 0 for D(Ma(x)). This setup makes D separate M outputs from Ma outputs, not source from target.
- [Section 4, Tables 1 and 2] The text in Section 4 states 'As in TJ-AIDL and MMFA, one of the datasets is used as the source domain and the other one is for target domain,' which confirms that Table 2 rows are not RAP-based. This discrepancy directly affects the central outperformance claim.
- [Section 4, Table 2 source/target design] The ambiguity between 'source = RAP' and 'source = Market/Duke with attributes' is not resolved anywhere in the text and is load-bearing for the state-of-the-art claim.
minor comments (4)
- [Section 3, Eq. (2)] Typographical error in the adversarial loss expression.
- [Section 3, paragraph after Eq. (3)] Ambiguity about which parameters are fixed during adaptation.
- [Section 4, Fig. 4] Figure clarity issue.
- [Section 4, hyper-parameter α] No ablation for α is reported.
Circularity Check
No circularity: the RAP-to-ReID adaptation is evaluated by before/after comparison, the hyperparameter is fixed, and the only self-citation is a non-load-bearing building block.
full rationale
The paper's central claim is that an attribute-recognition model trained on RAP can be adversarially adapted to an unlabeled ReID target, producing features useful for person re-identification. This is tested directly in Table 1 by comparing Rank1/mAP before adaptation (28.2%/8.7% and 15.6%/4.9%) with after adaptation (32.1%/10.6% and 18.7%/6.5%), so the main contribution is validated against its own baseline rather than by definition. The adversarial objective in Eq. (3), the attribute loss in Eq. (1), and the combined objective in Eq. (4) are stated explicitly and do not presuppose the reported ReID results. The hyperparameter alpha is fixed to 0.1 and is not fitted to the target metric. The self-citation to reference [14], authored by two of the present authors, is used only to justify a technical choice in the adaptation procedure (using samples from both domains and a least-squares adversarial loss); it is not the target result and does not by itself force the reported outcome. The apparent mismatch that Table 2 uses Market-1501 and DukeMTMC-reID as source domains rather than RAP is a protocol/external-validity concern, not a circularity: it does not make the prediction equivalent to its input. No step in the derivation reduces to a fitted parameter renamed as a prediction, and no uniqueness claim is imported from the authors' prior work. The derivation chain is therefore self-contained for the RAP-to-ReID experiments, and no circular step is present.
Assumptions & free parameters
free parameters (3)
- alpha =
0.1
- attribute subset =
70 selected from 91 in RAP
- learning rate =
0.0001
assumptions (3)
- domain assumption Attributes are consistent across different domains, so an attribute representation is transferable.
- domain assumption The least-squares adversarial objective in Eq. (3) trains a mapping M whose features cannot be distinguished from source features, yielding domain-invariant features.
- domain assumption The source attribute recognition model learned on RAP provides a reliable feature mapping Ma.
Cite this review
Pith. "Pith review of Unsupervised Domain-Adaptive Person Re-identification Based on Attributes." pith.science (2026). https://pith.science/paper/AFZ7IKGQ
@misc{pith2026190810359,
author = {Pith},
title = {Pith review of: Unsupervised Domain-Adaptive Person Re-identification Based on Attributes},
year = {2026},
howpublished = {\url{https://pith.science/paper/AFZ7IKGQ}},
note = {Machine review of arXiv:1908.10359}
}
read the original abstract
Pedestrian attributes, e.g., hair length, clothes type and color, locally describe the semantic appearance of a person. Training person re-identification (ReID) algorithms under the supervision of such attributes have proven to be effective in extracting local features which are important for ReID. Unlike person identity, attributes are consistent across different domains (or datasets). However, most of ReID datasets lack attribute annotations. On the other hand, there are several datasets labeled with sufficient attributes for the case of pedestrian attribute recognition. Exploiting such data for ReID purpose can be a way to alleviate the shortage of attribute annotations in ReID case. In this work, an unsupervised domain adaptive ReID feature learning framework is proposed to make full use of attribute annotations. We propose to transfer attribute-related features from their original domain to the ReID one: to this end, we introduce an adversarial discriminative domain adaptation method in order to learn domain invariant features for encoding semantic attributes. Experiments on three large-scale datasets validate the effectiveness of the proposed ReID framework.
Reference graph
Works this paper leans on
-
[1]
Unsupervised Domain-Adaptive Person Re-identification Based on Attributes
INTRODUCTION Person ReID aims at matching a target person across cameras. This is still a challenging problem due to large intra-personal variations and inter-personal similarities [1, 2]. For these rea- sons it is of paramount importance to have a model that gen- erates robust and consistent features modeling all these kind of variations. Pedestrian attr...
work page Pith review arXiv 1908
-
[2]
The effectiveness of using attributes for ReID has been reported in many works [1, 3, 4, 5, 16]
RELA TED WORKS This work mainly focuses on attribute based person ReID and unsupervised domain adaptation. The effectiveness of using attributes for ReID has been reported in many works [1, 3, 4, 5, 16]. Lin et al . jointly consider the ID and at- tribute as the supervisions in a deep multi-task model, to learn the discriminative ReID features [4]. Simila...
-
[3]
METHODOLOGY Suppose there are na samples from the attribute recognition (source) domainDa ={(xa i, ai)}na i=1, in which xa i represents theith sample, featuringm attributes ai∈ Rm. For the ReID (target) domainDp ={xp i}np i=1, there arenp samples xp i , but, on the contrary, no annotation is available. We want to learn a domain invariant mapping M, which ...
-
[4]
EXPERIMENTS In this section, experimental results are presented to demon- strate the effectiveness of the proposed ReID feature learning framework. To measure the performances of the proposed method, the Cumulative Matching Characteristic (CMC) and mean average precision (mAP) are used. Datasets: Three datasets, i.e., RAP, Market-1501 and Duke- MTMC-reID,...
-
[5]
CONCLUSIONS In this work, we presented an unsupervised domain adap- tive ReID framework for extracting attribute-related features. Considering that most of ReID datasets are not labeled with attributes, an unsupervised domain adaptation method has been proposed for learning domain invariant attribute-related features, which can be used for ReID in the tar...
-
[6]
Person re-identification: Past, present and future,
Liang Zheng, Yi Yang, and Alexander G Hauptmann, “Person re-identification: Past, present and future,” arXiv preprint arXiv:1610.02984, 2016
arXiv 2016
-
[7]
Exploiting gaussian mixture importance for person re-identification,
Xiangping Zhu, Amran Bhuiyan, Mohamed Lamine Mekhalfi, and Vittorio Murino, “Exploiting gaussian mixture importance for person re-identification,” in A VSS. IEEE, 2017, pp. 1–6
work page 2017
-
[8]
Person re-identification by attributes,
Ryan Layne, Timothy M Hospedales, Shaogang Gong, and Q Mary, “Person re-identification by attributes,” in BMVC, 2012, vol. 2, p. 8
work page 2012
Show all 28 references
-
[9]
Improving person re-identification by attribute and identity learning,
Yutian Lin, Liang Zheng, Zhedong Zheng, Yu Wu, and Yi Yang, “Improving person re-identification by attribute and identity learning,” arXiv preprint arXiv:1703.07220, 2017
2017 arXiv
-
[10]
Person re- identification using cnn features learned from combi- nation of attributes,
Tetsu Matsukawa and Einoshin Suzuki, “Person re- identification using cnn features learned from combi- nation of attributes,” in Pattern Recognition (ICPR), 2016 23rd International Conference on. IEEE, 2016, pp. 2428–2433
2016
-
[11]
Deepreid: Deep filter pairing neural network for per- son re-identification,
Wei Li, Rui Zhao, Tong Xiao, and Xiaogang Wang, “Deepreid: Deep filter pairing neural network for per- son re-identification,” in CVPR, 2014, pp. 152–159
2014
-
[12]
Mars: A video bench- mark for large-scale person re-identification,
Liang Zheng, Zhi Bie, Yifan Sun, Jingdong Wang, Chi Su, Shengjin Wang, and Qi Tian, “Mars: A video bench- mark for large-scale person re-identification,” in Euro- pean Conference on Computer Vision . Springer, 2016, pp. 868–884
2016
-
[13]
Person transfer gan to bridge domain gap for person re- identification,
Longhui Wei, Shiliang Zhang, Wen Gao, and Qi Tian, “Person transfer gan to bridge domain gap for person re- identification,” arXiv preprint arXiv:1711.08565, 2017
2017 arXiv
-
[14]
A richly annotated dataset for pedestrian attribute recognition,
Dangwei Li, Zhang Zhang, Xiaotang Chen, Haibin Ling, and Kaiqi Huang, “A richly annotated dataset for pedestrian attribute recognition,” arXiv preprint arXiv:1603.07054, 2016
2016 arXiv
-
[15]
Learning discrimi- native spatial representation for image classification,
Gaurav Sharma and Frederic Jurie, “Learning discrimi- native spatial representation for image classification,” in BMVC, 2011, pp. 1–11
2011
-
[16]
Human attribute recognition by deep hierarchical contexts,
Yining Li, Chen Huang, Chen Change Loy, and Xiaoou Tang, “Human attribute recognition by deep hierarchical contexts,” in European Conference on Computer Vision. Springer, 2016, pp. 684–700
2016
-
[17]
Scalable person re- identification: A benchmark,
Liang Zheng, Liyue Shen, Lu Tian, Shengjin Wang, Jingdong Wang, and Qi Tian, “Scalable person re- identification: A benchmark,” in ICCV, 2015, pp. 1116– 1124
2015
-
[18]
Adversarial discriminative domain adapta- tion,
Eric Tzeng, Judy Hoffman, Kate Saenko, and Trevor Darrell, “Adversarial discriminative domain adapta- tion,” in Computer Vision and Pattern Recognition (CVPR), 2017, vol. 1, p. 4
2017
-
[19]
Adversarial feature augmentation for unsupervised domain adaptation,
Riccardo V olpi, Pietro Morerio, Silvio Savarese, and Vittorio Murino, “Adversarial feature augmentation for unsupervised domain adaptation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 5495–5504
2018
-
[20]
Unla- beled samples generated by gan improve the person re- identification baseline in vitro,
Zhedong Zheng, Liang Zheng, and Yi Yang, “Unla- beled samples generated by gan improve the person re- identification baseline in vitro,” in Proceedings of the IEEE Conference on Computer Vision and Computer Vi- sion, 2017, pp. 3754–3762
2017
-
[21]
Deep attributes driven multi-camera person re-identification,
Chi Su, Shiliang Zhang, Junliang Xing, Wen Gao, and Qi Tian, “Deep attributes driven multi-camera person re-identification,” in European conference on computer vision. Springer, 2016, pp. 475–491
2016
-
[22]
Transferable joint attribute-identity deep learning for unsupervised person re-identification,
Jingya Wang, Xiatian Zhu, Shaogang Gong, and Wei Li, “Transferable joint attribute-identity deep learning for unsupervised person re-identification,” CVPR2018, 2018
2018
-
[23]
Multi-task mid-level feature alignment network for unsupervised cross-dataset person re-identification,
Shan Lin, Haoliang Li, Chang-Tsun Li, and Alex Chichung Kot, “Multi-task mid-level feature alignment network for unsupervised cross-dataset person re-identification,” BMVC2018, 2018
2018
-
[24]
Unsupervised domain adaptation by backpropagation,
Yaroslav Ganin and Victor Lempitsky, “Unsupervised domain adaptation by backpropagation,” in Interna- tional Conference on Machine Learning , 2015, pp. 1180–1189
2015
-
[25]
Domain adaptation for visual ap- plications: A comprehensive survey,
Gabriela Csurka, “Domain adaptation for visual ap- plications: A comprehensive survey,” arXiv preprint arXiv:1702.05374, 2017
2017 arXiv
-
[26]
Minimal-entropy correlation alignment for unsuper- vised deep domain adaptation,
Pietro Morerio, Jacopo Cavazza, and Vittorio Murino, “Minimal-entropy correlation alignment for unsuper- vised deep domain adaptation,” International Confer- ence on Learning Representations, 2018
2018
-
[27]
Generative adversarial nets,
Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio, “Generative adversarial nets,” in NIPS, 2014, pp. 2672–2680
2014
-
[28]
Least squares generative adversarial networks,
Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley, “Least squares generative adversarial networks,” in ICCV. IEEE, 2017, pp. 2813–2821
2017
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.