Pith. sign in

REVIEW 3 major objections 4 minor 38 references

Open-Attribute Recognition for Person Retrieval: Finding People Through Distinctive and Novel Attributes

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper proposes the Open-Attribute Recognition for Person Retrieval (OAPR) task, in which a person is retrieved by attribute cues regardless of whether those attributes appeared in training.

desk verdict Proposes a genuinely useful task framing for open-vocabulary person retrieval, but the abstract alone can't support the empirical claims, and the dataset split design is the load-bearing detail to check. read the letter →

arxiv 2508.01389 v2 pith:KL5RX37E submitted 2025-08-02 cs.CV

Minjeong Park , Hongbeen Park , Sangwon Lee , Yoonha Jang , Jinkyu Kim This is my paper
classification cs.CV
keywords open-attributerecognitionpersonretrievalpedestrianattributebodypartrepresentationsgeneralizableattribute-basedseen/unseensplitbenchmarkreconstruction
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

Pedestrian attribute recognition conventionally assumes a closed set: every attribute class present at inference was seen during training. The paper argues this is unrealistic for person retrieval, because new attributes can emerge in the wild and because benchmark attributes are often generic and shared across people, so they cannot single out a target. It therefore proposes Open-Attribute Recognition for Person Retrieval (OAPR), where the query is any attribute description, seen or unseen. To support OAPR, it introduces a framework that learns generalizable body-part representations spanning a broad range of attribute categories, and it rebuilds four standard datasets into seen/unseen attribute splits for evaluation. The paper's experiments are offered as evidence that closed-set methods are inadequate for this task and that the proposed framework is effective at it.

What carries the argument

The central mechanism is the paper's framework for learning generalizable body-part representations. Instead of tying recognition to a fixed list of attribute classes, the framework decomposes a person into body-part regions and learns a representation space in which visual features of those parts can be matched against attribute descriptions, including ones not in the training vocabulary. This part-level matching is what lets an unseen attribute query be evaluated by composing the relevant body-part cues rather than by looking up a classifier. The four reconstructed datasets provide the seen/unseen attribute protocol under which this mechanism is measured.

What would settle it

Build a retrieval test whose query attributes are excluded not just from training but from the entire benchmark vocabulary, with no visual or linguistic overlap with any seen attribute; if ranking accuracy collapses on those queries while remaining high on held-out benchmark attributes, the claim that the framework handles novel attributes would be refuted.

Watch

Extended reading notes

Core claim

The core claim is that a person-retrieval model trained on a fixed attribute vocabulary can still rank people correctly for attribute descriptions it never saw, provided it represents people at the level of body parts rather than as a set of predefined attribute classifiers. On the paper's terms, this means the model learns visual representations that are general enough to cover a broad range of attribute categories, so a query such as a rarely used or newly phrased attribute can be matched to the right person without retraining. The paper further claims that current closed-set PAR methods have a capability gap: they cannot absorb novel attributes without changing the model's output vocabulary, which is exactly the limitation OAPR is designed to remove. The reconstructed datasets make this capability measurable by holding out a subset of attributes during training and testing retrieval on those held-out cues.

Load-bearing premise

The load-bearing premise is that the held-out attributes in the reconstructed benchmarks stand in for truly novel attribute descriptions; if those held-out attributes are visually or linguistically close to training attributes, the reported results would show interpolation within a known semantic space rather than handling of genuinely new attributes.

Editorial extensions

If this is right

  • Closed-set pedestrian attribute recognition is insufficient for real-world person retrieval, because new attribute descriptions cannot be added without retraining the classifier head.
  • A model with generalizable body-part representations can answer queries in attribute vocabulary it never saw, so retrieval systems can be deployed with flexible, user-generated attribute descriptions.
  • Distinctive and novel attributes are more useful for disambiguating a target person than the generic, shared attributes found in standard benchmarks.
  • The four reconstructed datasets give the community a common evaluation protocol for open-attribute retrieval, making the capability gap between closed-set and open-attribute methods measurable.

Reading between the lines

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

  • If the part-level representation space is as general as claimed, an obvious next step is compositional queries: unseen attribute phrases that combine familiar part concepts should be retrievable without any training examples, a test the paper does not run.
  • The benchmark split treats held-out attributes as proxies for true novelty; a stronger stress test would use attributes that share no visual or linguistic structure with the training vocabulary, since strong performance there would show genuine extrapolation rather than interpolation within a known attribute space.
  • The open-attribute protocol could also serve as a diagnosis tool for vision-language models, measuring directly how much they overfit common attribute names when retrieving people on the reconstructed seen/unseen splits.
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

3 major / 4 minor

Summary. The paper proposes a new task, Open-Attribute Recognition for Person Retrieval (OAPR), in which a model must retrieve people from attribute cues even when the attribute was not present in the training vocabulary. To address this, the authors present a framework that learns generalizable body-part representations and reconstruct four widely used pedestrian attribute datasets to create seen/unseen attribute splits. The abstract claims that comprehensive experiments demonstrate the necessity of the task and the effectiveness of the framework, but it reports no quantitative results. The supplied full text is largely illegible due to severe character corruption, so this review is based primarily on the abstract, the recognizable section structure, and the visible limitation statements; the experimental tables and equations could not be verified.

Significance. If the claims are substantiated, the paper addresses a real and practical limitation of closed-set attribute-based person retrieval: existing methods fail when new attribute vocabulary appears after deployment. A model that can rank people by never-seen attribute descriptions, together with reproducible benchmark protocols, would be a useful contribution to the computer vision community. The authors commit to releasing source code and pre-trained models, which is a clear strength. However, the significance of the work depends entirely on the empirical evidence that the 'unseen' attributes are genuinely novel rather than compositions of already-seen concepts; this evidence is not visible in the abstract and could not be verified in the corrupted full text.

major comments (3)
  1. [Section 4 (Experiments, dataset reconstruction)] The load-bearing assumption of the OAPR evaluation is that the held-out attributes are genuinely unseen rather than merely labels held out from the same closed vocabulary. The abstract states that the authors reconstruct four widely used datasets, but it does not describe how the seen/unseen split is defined. If the unseen attributes are selected from the same attribute vocabulary as the training attributes, then a model that combines known body-part concepts (for example, 'black' and 'backpack') can answer such queries without exhibiting open-set generalization. The authors must provide a quantitative characterization of the semantic distance between the seen and unseen attribute sets, or evaluate on attributes that are not in the original benchmark vocabulary at all, for the central claim to hold.
  2. [Abstract and Section 4 (Experiments)] The abstract claims that 'comprehensive experiments ... demonstrate the necessity of the OAPR task and the effectiveness of our framework,' but it reports no numbers, baseline comparisons, or ablations. In the supplied full text, the tables are not legible, so I could not verify the experimental results. Even if the full text contains these results, the paper should state at least the key metric (for example, mAP or Rank-1) on the four datasets and include a comparison with closed-set PAR baselines under the same protocol. Without such a reference point, the claimed 'necessity' of the OAPR task is not falsifiable from the abstract alone.
  3. [Limitations section (near the end of the manuscript)] The appended limitation statements appear to acknowledge that the reconstructed datasets are based on existing benchmarks and inherit their attribute vocabularies. This is directly relevant to the claim of handling novel attributes: if the unseen attributes are drawn from the same vocabulary, the evaluation may show interpolation within a known semantic space rather than generalization to truly novel attribute descriptions. The authors should explicitly address this limitation in the main text and provide a direct experiment with attributes not appearing anywhere in the benchmark vocabulary, or with newly collected attribute descriptions, to support the open-attribute claim.
minor comments (4)
  1. [Abstract] The abstract should name the four reconstructed datasets and state the evaluation metric; currently the reader must infer the scale of the experiments from the phrase 'comprehensive experiments,' which is too vague for a claims-heavy abstract.
  2. [Title and contribution list] The wording is inconsistent about whether the task is 'attribute recognition' or 'person retrieval'; since OAPR is a retrieval/ranking task, the terminology should be aligned throughout the paper.
  3. [Full text (PDF rendering)] The manuscript text supplied to the reviewer contains extensive character corruption that prevents verification of equations, tables, and precise wording of the limitation statements; the authors should ensure that the uploaded PDF is not corrupted and that all tables and equations are legible.
  4. [Section 1 (Introduction) / Section 3 (Method)] The terms 'attribute cues,' 'attribute categories,' and 'attributes' appear to be used interchangeably; the authors should define these terms at first use, especially in the description of how seen and unseen attribute splits are constructed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the OAPR claim rests on empirical dataset experiments, not on a derivation that reduces to its own inputs.

full rationale

The abstract's central assertion, that OAPR retrieves people from attributes regardless of whether those attributes were seen in training, is supported by a proposed framework and reconstructed datasets. No equation, fitted-parameter-then-prediction step, or load-bearing self-citation appears in the abstract that would make the conclusion equivalent to an input. The statement 'comprehensive experiments on these datasets demonstrate the necessity of the OAPR task and the effectiveness of our framework' is an empirical evaluation claim, not a circular derivation: proposing a benchmark protocol does not, by itself, define the target result in terms of the benchmark construction. The concern that held-out 'unseen' attributes may still belong to the same closed vocabulary is a possible external-validity limitation, but it is not a demonstrated circular step because no quoted text shows that the split was constructed from the model's outputs or that the evaluation metric is the training objective. The supplied full text is largely encoding-corrupted, and the abstract contains no derivational chain; under the rule that circularity must be exhibited with a specific quote and reduction, no such step can be identified. The appropriate finding is therefore no significant circularity.

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

Ledger entries are reconstructed from the abstract because the full text is unreadable. The central empirical claim depends on the model's transferability premise (axiom 1), on the fairness of the author-chosen unseen attribute splits (axiom 2 and the split-proportion free parameter), and on the correctness of the underlying dataset annotations (axiom 3). No new physical or mathematical entities are proposed; the only self-created evaluation substrate is the set of four reconstructed benchmarks, which has no external validation visible in the abstract. A full audit would be needed on the actual manuscript to name architecture-specific hyperparameters (loss weights, part-embedding dimensions), which cannot be enumerated here.

free parameters (1)
  • Unseen attribute split proportion per dataset = not stated in abstract
    The authors decide which attribute classes are held out as 'unseen' in each reconstructed dataset; this hand-chosen protocol constant controls the difficulty of the open-attribute task and is invisible at the abstract level.
assumptions (3)
  • domain assumption Body-part representations trained on seen attribute categories transfer to unseen attribute categories at query time.
    This transferability premise is the load-bearing assumption of the proposed framework; it is asserted in the abstract ('generalizable body part representations') and is validated only by the paper's own experiments, which cannot be inspected here.
  • domain assumption Held-out 'unseen' attribute splits in the reconstructed datasets are representative of novel attributes that would emerge in real deployments.
    The validity of the OAPR protocol rests on the unseen attribute splits being a fair simulation of genuinely novel attribute vocabulary; the split-construction procedure is not visible at the abstract level.
  • domain assumption Attribute annotations in the four reconstructed datasets are correct and consistent enough to serve as ground truth for retrieval evaluation.
    PAR ground-truth annotations are known to be noisy; the reconstruction process presumably filters or merges labels, but this is invisible in the abstract.
invented entities (1)
  • OAPR benchmark reconstructions of four PAR datasets (dataset names not listed in the abstract)
    purpose: Provide the evaluation substrate for the claimed open-attribute retrieval capability; seen/unseen attribute splits are defined by the authors.
    The benchmarks are constructed by the authors and their protocols are not externally anchored in the abstract; if the splits are unrepresentative, the central empirical claim lacks independent support.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Open-Attribute Recognition for Person Retrieval: Finding People Through Distinctive and Novel Attributes." pith.science (2026). https://pith.science/paper/KL5RX37E

@misc{pith2026250801389,
  author       = {Pith},
  title        = {Pith review of: Open-Attribute Recognition for Person Retrieval: Finding People Through Distinctive and Novel Attributes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KL5RX37E}},
  note         = {Machine review of arXiv:2508.01389}
}
read the original abstract

Pedestrian Attribute Recognition (PAR) plays a crucial role in various vision tasks such as person retrieval and identification. Most existing attribute-based retrieval methods operate under the closed-set assumption that all attribute classes are consistently available during both training and inference. However, this assumption limits their applicability in real-world scenarios where novel attributes may emerge. Moreover, predefined attributes in benchmark datasets are often generic and shared across individuals, making them less discriminative for retrieving the target person. To address these challenges, we propose the Open-Attribute Recognition for Person Retrieval (OAPR) task, which aims to retrieve individuals based on attribute cues, regardless of whether those attributes were seen during training. To support this task, we introduce a novel framework designed to learn generalizable body part representations that cover a broad range of attribute categories. Furthermore, we reconstruct four widely used datasets for open-attribute recognition. Comprehensive experiments on these datasets demonstrate the necessity of the OAPR task and the effectiveness of our framework. The source code and pre-trained models will be publicly available upon publication.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 30 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Chen, W.; Xu, X.; Jia, J.; Luo, H.; Wang, Y.; Wang, F.; Jin, R.; and Sun, X. 2023. Beyond Appearance: a Semantic Controllable Self-Supervised Learning Framework for Human-Centric Visual Tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 15050--15061

    work page 2023

  4. [4]

    Chen, Y.; Zhang, G.; Lu, Y.; Wang, Z.; and Zheng, Y. 2022. TIPCB: A simple but effective part-based convolutional baseline for text-based person search. Neurocomputing, 494: 171--181

    work page 2022

  5. [5]

    Cheng, T.; Song, L.; Ge, Y.; Liu, W.; Wang, X.; and Shan, Y. 2024. Yolo-world: Real-time open-vocabulary object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 16901--16911

    work page 2024

  6. [6]

    Cheng, X.; Jia, M.; Wang, Q.; and Zhang, J. 2022. A simple visual-textual baseline for pedestrian attribute recognition. IEEE Transactions on Circuits and Systems for Video Technology, 32(10): 6994--7004

    work page 2022

  7. [7]

    C.; and Tang, X

    Deng, Y.; Luo, P.; Loy, C. C.; and Tang, X. 2014. Pedestrian attribute recognition at far distance. In Proceedings of the 22nd ACM international conference on Multimedia, 789--792

    work page 2014

  8. [8]

    Devlin, J.; Chang, M.-W.; Lee, K.; and Toutanova, K. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805

    arXiv 2018

Show all 38 references
  1. [9]

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

    2020 arXiv

  2. [10]

    Ergasti, A.; Fontanini, T.; Ferrari, C.; Bertozzi, M.; and Prati, A. 2024. MARS: Paying more attention to visual attributes for text-based person search. arXiv preprint arXiv:2407.04287

    2024 arXiv

  3. [11]

    He, S.; Luo, H.; Jiang, W.; Jiang, X.; and Ding, H. 2023. VGSG: Vision-guided semantic-group network for text-based person search. IEEE Transactions on Image Processing, 33: 163--176

    2023

  4. [12]

    Huang, Y.; Zhang, Z.; Wu, Q.; Zhong, Y.; and Wang, L. 2024 a . Attribute-guided pedestrian retrieval: Bridging person re-id with internal attribute variability. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 17689--17699

    2024

  5. [13]

    Huang, Z.; Wu, X.; Chen, X.; Zhao, H.; Zhu, L.; and Lasenby, J. 2024 b . Openins3d: Snap and lookup for 3d open-vocabulary instance segmentation. In European Conference on Computer Vision, 169--185. Springer

    2024

  6. [14]

    Jeong, B.; Park, J.; and Kwak, S. 2021. Asmr: Learning attribute-based person search with adaptive semantic margin regularizer. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 12016--12025

    2021

  7. [15]

    Jia, J.; Huang, H.; Chen, X.; and Huang, K. 2021. Rethinking of Pedestrian Attribute Recognition: A Reliable Evaluation under Zero-Shot Pedestrian Identity Setting. arXiv preprint arXiv:2107.03576

    2021 arXiv

  8. [16]

    Jiang, D.; and Ye, M. 2023. Cross-modal implicit relation reasoning and aligning for text-to-image person retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2787--2797

    2023

  9. [17]

    U.; Rasheed, H.; Maaz, M.; Khan, S.; and Khan, F

    Khattak, M. U.; Rasheed, H.; Maaz, M.; Khan, S.; and Khan, F. S. 2023. MaPLe: Multi-Modal Prompt Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 19113--19122

    2023

  10. [18]

    Li, D.; Chen, X.; and Huang, K. 2015. Multi-attribute Learning for Pedestrian Attribute Recognition in Surveillance Scenarios. In ACPR, 111--115

    2015

  11. [19]

    Li, D.; Zhang, Z.; Chen, X.; and Huang, K. 2019. A richly annotated pedestrian dataset for person retrieval in real surveillance scenarios. IEEE transactions on image processing, 28(4): 1575--1590

    2019

  12. [20]

    Li, S.; Xiao, T.; Li, H.; Zhou, B.; Yue, D.; and Wang, X. 2017. Person search with natural language description. In Proceedings of the IEEE conference on computer vision and pattern recognition, 1970--1979

    2017

  13. [21]

    Li, Y.; Wang, H.; Duan, Y.; and Li, X. 2023. CLIP Surgery for Better Explainability with Enhancement in Open-Vocabulary Tasks. arXiv:2304.05653

    2023 arXiv

  14. [22]

    Lin, Y.; Zheng, L.; Zheng, Z.; Wu, Y.; Hu, Z.; Yan, C.; and Yang, Y. 2019. Improving person re-identification by attribute and identity learning. Pattern Recognition, 95: 151–161

    2019

  15. [23]

    Liu, X.; Zhao, H.; Tian, M.; Sheng, L.; Shao, J.; Yan, J.; and Wang, X. 2017. HydraPlus-Net: Attentive Deep Features for Pedestrian Analysis. In Proceedings of the IEEE international conference on computer vision, 1--9

    2017

  16. [24]

    Park, M.; Park, H.; and Kim, J. 2025. ViTA-PAR: Visual and Textual Attribute Alignment with Attribute Prompting for Pedestrian Attribute Recognition

    2025

  17. [25]

    Paszke, A.; Gross, S.; Massa, F.; Lerer, A.; Bradbury, J.; Chanan, G.; Killeen, T.; Lin, Z.; Gimelshein, N.; Antiga, L.; Desmaison, A.; Kopf, A.; Yang, E.; DeVito, Z.; Raison, M.; Tejani, A.; Chilamkurthy, S.; Steiner, B.; Fang, L.; Bai, J.; and Chintala, S. 2019. PyTorch: An ...

    2019

  18. [26]

    Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V.; Vanderplas, J.; Passos, A.; Cournapeau, D.; Brucher, M.; Perrot, M.; and Duchesnay, E. 2011. Scikit-learn: Machine Learning in P ython. Journ...

    2011

  19. [27]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, 8748--8763. PMLR

    2021

  20. [28]

    Reimers, N.; and Gurevych, I. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics

    2019

  21. [29]

    Wang, S.; Duan, Y.; Ding, H.; Tan, Y.-P.; Yap, K.-H.; and Yuan, J. 2022 a . Learning transferable human-object interaction detector with natural language supervision. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 939--948

    2022

  22. [30]

    Wang, X.; Jin, J.; Li, C.; Tang, J.; Zhang, C.; and Wang, W. 2025. Pedestrian Attribute Recognition via CLIP-Based Prompt Vision-Language Fusion. IEEE Transactions on Circuits and Systems for Video Technology

    2025

  23. [31]

    Wang, Z.; Zhu, A.; Xue, J.; Wan, X.; Liu, C.; Wang, T.; and Li, Y. 2022 b . Caibc: Capturing all-round information beyond color for text-based person retrieval. In Proceedings of the 30th ACM international conference on multimedia, 5314--5322

    2022

  24. [32]

    Wu, P.; Zhou, X.; Pang, G.; Sun, Y.; Liu, J.; Wang, P.; and Zhang, Y. 2024. Open-vocabulary video anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 18297--18307

    2024

  25. [33]

    Zhai, Y.; Zeng, Y.; Huang, Z.; Qin, Z.; Jin, X.; and Cao, D. 2024. Multi-Prompts Learning with Cross-Modal Alignment for Attribute-Based Person Re-identification. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 6979--6987

    2024

  26. [34]

    Zhang, Y.; Wang, S.; Kan, S.; Weng, Z.; Cen, Y.; and Tan, Y. 2023. POAR: Towards Open Vocabulary Pedestrian Attribute Recognition. In Proceedings of the 31st ACM International Conference on Multimedia (MM '23)

    2023

  27. [35]

    C.; and Liu, Z

    Zhou, K.; Yang, J.; Loy, C. C.; and Liu, Z. 2022 a . Conditional Prompt Learning for Vision-Language Models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2022

  28. [36]

    C.; and Liu, Z

    Zhou, K.; Yang, J.; Loy, C. C.; and Liu, Z. 2022 b . Learning to Prompt for Vision-Language Models. International Journal of Computer Vision (IJCV)

    2022

  29. [37]

    Zhou, Q.; Pang, G.; Tian, Y.; He, S.; and Chen, J. 2023. AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection. In The Twelfth International Conference on Learning Representations

    2023

  30. [38]

    Zhu, J.; Liu, L.; Zhan, Y.; Zhu, X.; Zeng, H.; and Tao, D. 2023. Attribute-image person re-identification via modal-consistent metric learning. International Journal of Computer Vision, 131(11): 2959--2976

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.