pith. sign in

arxiv: 2606.11626 · v1 · pith:5DWOT3DNnew · submitted 2026-06-10 · 💻 cs.CV

Adapting Vision-Language Models from Iconic to Inclusive for Multi-Label Recognition Without Labels

Cheng Chen , Jingyu Zhou , Yifan Zhao , Jia Li This is my paper

Pith reviewed 2026-06-27 10:26 UTC · model grok-4.3

classification 💻 cs.CV
keywords vision-language modelsmulti-label recognitionunsupervised adaptationzero-shot learningimage recognitionmodel adaptationinclusive understanding
0
0 comments X

The pith

Vision-language models can be adapted without labels to recognize multiple objects by shifting from iconic single-object focus to inclusive multi-label understanding.

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

The paper aims to fix the tendency of vision-language models to latch onto the single most prominent object in an image and ignore others. This bias prevents effective multi-label recognition where several objects or attributes matter. The authors introduce an unsupervised two-stage process: a cutting stage that uses multi-sampling to spread model attention, followed by a sewing stage that blends adaptations to match multi-label distributions while retaining the model's original behavior after one epoch. If the approach holds, pre-trained models could deliver stronger multi-label results on public datasets without any manual labels, outperforming prior unsupervised methods and some weakly supervised ones.

Core claim

The central claim is that the proposed cutting and sewing framework adapts vision-language models from iconic to inclusive recognition, enabling label-free multi-label image recognition that outperforms existing unsupervised methods on four public datasets and even surpasses several representative weakly supervised baselines.

What carries the argument

The two-stage cutting and sewing process, consisting of the multi-sampling response estimator to avoid single-object concentration and the multi-object blend adaptation to align outputs with multi-label distributions while preserving original model traits in one epoch.

Load-bearing premise

The multi-sampling response estimator and multi-object blend adaptation can shift the model toward multi-label behavior while preserving original characteristics in only one epoch without introducing new biases or degrading single-object performance.

What would settle it

If the adapted model fails to detect secondary objects in multi-label test images or shows clear drops in single-object zero-shot accuracy on the four datasets, the adaptation claim would be falsified.

read the original abstract

Understanding multi-label images remains a challenging task in computer vision. With the rapid progress of vision-language multimodal learning, vision-language models (VLMs) enable zero-shot recognition without labeled data. However, due to their intrinsic design, these models often prioritize the most iconic object and omit other contextual positives. This intrinsic bias conflicts with the nature of multi-label learning, thereby limiting their applicability. In this work, we propose an unsupervised framework that adapts VLMs from iconic recognition toward inclusive understanding, enabling label-free multi-label image recognition. Our approach consists of two key stages, ``cutting'' and ``sewing'': In the cutting stage, we present the multi-sampling response estimator to prevent the model from concentrating only on one single object. In the second sewing stage, the multi-object blend adaptation is introduced to adjust the labels to better conform to the multi-label distribution while preserving the intrinsic characteristics of the original model within only one epoch. Extensive experiments show that our framework significantly outperforms existing unsupervised approaches on four public datasets, even surpassing several representative weakly supervised baselines. These results demonstrate the potential of adapting pre-trained VLMs for more comprehensive visual understanding without manual annotations. Our code is publicly available at https://github.com/iCVTEAM/TailorCLIP.

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

2 major / 2 minor

Summary. The paper claims to introduce an unsupervised two-stage framework (TailorCLIP) that adapts pre-trained vision-language models from iconic single-object recognition to inclusive multi-label recognition without any labels. The cutting stage uses a multi-sampling response estimator to avoid single-object focus; the sewing stage applies multi-object blend adaptation in a single epoch to align label distributions with multi-label statistics while preserving the original model's characteristics. Experiments on four public datasets reportedly show gains over unsupervised baselines and parity or superiority to some weakly-supervised methods.

Significance. If the one-epoch adaptation genuinely produces inclusive multi-label behavior while exactly retaining single-object decision boundaries and introducing no systematic biases, the result would be significant for efficient, label-free extension of VLMs to multi-label tasks. Public code release strengthens reproducibility. The significance hinges on whether the reported gains can be attributed to the proposed mechanism rather than unverified assumptions about preservation.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (sewing stage): the claim that multi-object blend adaptation 'adjust[s] the labels to better conform to the multi-label distribution while preserving the intrinsic characteristics of the original model' is load-bearing for attributing gains to genuine inclusive adaptation, yet no quantitative check (e.g., single-object accuracy or decision-boundary comparison before vs. after the epoch) is described to test retention or absence of new biases.
  2. [§4] §4 (experiments): the assertion that the framework 'significantly outperforms existing unsupervised approaches... even surpassing several representative weakly supervised baselines' cannot be evaluated without reported details on baseline re-implementations, exact dataset splits, or error bars; the one-epoch claim therefore rests on unverified experimental support.
minor comments (2)
  1. [§3.1] Notation for the multi-sampling response estimator is introduced without an explicit equation or pseudocode; adding one would clarify the cutting stage.
  2. [§4] The four datasets are named but their multi-label statistics (e.g., average labels per image) are not tabulated; this would help interpret the adaptation target.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful comments on our manuscript. We address each major point below with clarifications on our design choices and experimental reporting. Where additional verification strengthens the claims, we will incorporate it in revision.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (sewing stage): the claim that multi-object blend adaptation 'adjust[s] the labels to better conform to the multi-label distribution while preserving the intrinsic characteristics of the original model' is load-bearing for attributing gains to genuine inclusive adaptation, yet no quantitative check (e.g., single-object accuracy or decision-boundary comparison before vs. after the epoch) is described to test retention or absence of new biases.

    Authors: The single-epoch constraint and blend formulation are explicitly designed to limit parameter drift and retain the original decision boundaries, as the adaptation operates only on blended pseudo-labels without full fine-tuning. We acknowledge that an explicit before/after quantitative check on single-object accuracy would provide stronger evidence. In the revised version we will add such a comparison (e.g., zero-shot single-label accuracy on ImageNet subsets before and after the sewing stage) to directly verify preservation. revision: yes

  2. Referee: [§4] §4 (experiments): the assertion that the framework 'significantly outperforms existing unsupervised approaches... even surpassing several representative weakly supervised baselines' cannot be evaluated without reported details on baseline re-implementations, exact dataset splits, or error bars; the one-epoch claim therefore rests on unverified experimental support.

    Authors: All baselines were re-implemented from their public code using the official dataset splits cited in the respective papers (MS-COCO 2014 train/val, VOC 2007, etc.). Results are averaged over three random seeds with standard deviation reported in the tables. We will expand §4 with a dedicated paragraph detailing the exact re-implementation settings, splits, and seed values to make the one-epoch results fully reproducible. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the adaptation framework

full rationale

The paper introduces a novel two-stage unsupervised adaptation (cutting via multi-sampling response estimator, sewing via multi-object blend adaptation) presented as an original construction to shift VLMs toward multi-label behavior. No equations, fitted parameters, or self-citations are shown reducing the central claims to inputs by definition or construction. Performance gains are reported as empirical results on public datasets, independent of any load-bearing self-referential derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only; the framework rests on the domain assumption that VLMs exhibit a measurable iconic bias that can be corrected by sampling and blending without external labels. No free parameters or invented entities are named in the provided text.

axioms (1)
  • domain assumption VLMs intrinsically prioritize the most iconic object and omit contextual positives in multi-label images
    Explicitly stated in the abstract as the core limitation being addressed.

pith-pipeline@v0.9.1-grok · 5758 in / 1101 out tokens · 24963 ms · 2026-06-27T10:26:32.809695+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

79 extracted references · 3 canonical work pages · 1 internal anchor

  1. [1]

    FirstName LastName , title =

  2. [2]

    FirstName Alpher , title =

  3. [3]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

  4. [4]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

  5. [5]

    FirstName Alpher and FirstName Gamow , title =

  6. [6]

    2016 , volume =

    He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian , booktitle = CVPR, title =. 2016 , volume =

    2016

  7. [7]

    2023 , pages =

    Pathiraja, Bimsara and Gunawardhana, Malitha and Khan, Muhammad Haris , title =. 2023 , pages =

    2023

  8. [8]

    2023 , pages =

    Wu, Size and Zhang, Wenwei and Jin, Sheng and Liu, Wentao and Loy, Chen Change , title =. 2023 , pages =

    2023

  9. [9]

    2023 , pages =

    Ma, Yuqing and Li, Hainan and Zhang, Zhange and Guo, Jinyang and Zhang, Shanghang and Gong, Ruihao and Liu, Xianglong , title =. 2023 , pages =

    2023

  10. [10]

    2023 , pages =

    Liu, Chang and Zhang, Weiming and Lin, Xiangru and Zhang, Wei and Tan, Xiao and Han, Junyu and Li, Xiaomao and Ding, Errui and Wang, Jingdong , title =. 2023 , pages =

    2023

  11. [11]

    2023 , pages =

    Ding, Jian and Xue, Nan and Xia, Gui-Song and Schiele, Bernt and Dai, Dengxin , title =. 2023 , pages =

    2023

  12. [12]

    2023 , pages =

    Zhao, Zhen and Yang, Lihe and Long, Sifan and Pi, Jimin and Zhou, Luping and Wang, Jingdong , title =. 2023 , pages =

    2023

  13. [13]

    2023 , pages =

    Liu, Xinyu and Tian, Beiwen and Wang, Zhen and Wang, Rui and Sheng, Kehua and Zhang, Bo and Zhao, Hao and Zhou, Guyue , title =. 2023 , pages =

    2023

  14. [14]

    2023 , pages =

    Li, Ruihuang and He, Chenhang and Li, Shuai and Zhang, Yabin and Zhang, Lei , title =. 2023 , pages =

    2023

  15. [15]

    2023 , pages =

    Lin, Yuqi and Chen, Minghao and Wang, Wenxiao and Wu, Boxi and Li, Ke and Lin, Binbin and Liu, Haifeng and He, Xiaofei , title =. 2023 , pages =

    2023

  16. [16]

    2023 , pages =

    Zhou, Ziqin and Lei, Yinjie and Zhang, Bowen and Liu, Lingqiao and Liu, Yifan , title =. 2023 , pages =

    2023

  17. [17]

    2023 , pages =

    Zeng, Zequn and Zhang, Hao and Lu, Ruiying and Wang, Dongsheng and Chen, Bo and Wang, Zhengjue , title =. 2023 , pages =

    2023

  18. [18]

    2023 , pages =

    Ramos, Rita and Martins, Bruno and Elliott, Desmond and Kementchedjhieva, Yova , title =. 2023 , pages =

    2023

  19. [19]

    2023 , pages =

    Lee, Seongwon and Lee, Suhyeon and Seong, Hongje and Kim, Euntai , title =. 2023 , pages =

    2023

  20. [20]

    2023 , pages =

    Sain, Aneeshan and Bhunia, Ayan Kumar and Chowdhury, Pinaki Nath and Koley, Subhadeep and Xiang, Tao and Song, Yi-Zhe , title =. 2023 , pages =

    2023

  21. [21]

    2023 , pages =

    Xie, Chen-Wei and Sun, Siyang and Xiong, Xiong and Zheng, Yun and Zhao, Deli and Zhou, Jingren , title =. 2023 , pages =

    2023

  22. [22]

    2023 , pages =

    Saito, Kuniaki and Sohn, Kihyuk and Zhang, Xiang and Li, Chun-Liang and Lee, Chen-Yu and Saenko, Kate and Pfister, Tomas , title =. 2023 , pages =

    2023

  23. [23]

    2023 , pages =

    Kim, Youngwook and Kim, Jae Myung and Jeong, Jieun and Schmid, Cordelia and Akata, Zeynep and Lee, Jungwoo , title =. 2023 , pages =

    2023

  24. [24]

    2023 , pages =

    Guo, Zixian and Dong, Bowen and Ji, Zhilong and Bai, Jinfeng and Guo, Yiwen and Zuo, Wangmeng , title =. 2023 , pages =

    2023

  25. [25]

    2023 , pages =

    Gupta, Rohit and Roy, Anirban and Christensen, Claire and Kim, Sujeong and Gerard, Sarah and Cincebeaux, Madeline and Divakaran, Ajay and Grindal, Todd and Shah, Mubarak , title =. 2023 , pages =

    2023

  26. [26]

    2023 , pages =

    Liu, Ziming and Guo, Song and Lu, Xiaocheng and Guo, Jingcai and Zhang, Jiewei and Zeng, Yue and Huo, Fushuo , title =. 2023 , pages =

    2023

  27. [27]

    2023 , pages =

    Zhang, Wenqiao and Liu, Changshuo and Zeng, Lingze and Ooi, Bengchin and Tang, Siliang and Zhuang, Yueting , title =. 2023 , pages =

    2023

  28. [28]

    2023 , pages =

    Zhu, Ke and Fu, Minghao and Wu, Jianxin , title =. 2023 , pages =

    2023

  29. [29]

    2023 , pages =

    Zhu, Xuelin and Liu, Jian and Liu, Weijia and Ge, Jiawei and Liu, Bo and Cao, Jiuxin , title =. 2023 , pages =

    2023

  30. [30]

    2023 , pages =

    Xia, Xiaobo and Deng, Jiankang and Bao, Wei and Du, Yuxuan and Han, Bo and Shan, Shiguang and Liu, Tongliang , title =. 2023 , pages =

    2023

  31. [31]

    2023 , pages =

    Li, Miaoge and Wang, Dongsheng and Liu, Xinyang and Zeng, Zequn and Lu, Ruiying and Chen, Bo and Zhou, Mingyuan , title =. 2023 , pages =

    2023

  32. [32]

    2023 , pages =

    Yang, Penghui and Xie, Ming-Kun and Zong, Chen-Chen and Feng, Lei and Niu, Gang and Sugiyama, Masashi and Huang, Sheng-Jun , title =. 2023 , pages =

    2023

  33. [33]

    2023 , pages =

    Dong, Songlin and Luo, Haoyu and He, Yuhang and Wei, Xing and Cheng, Jie and Gong, Yihong , title =. 2023 , pages =

    2023

  34. [34]

    2023 , pages =

    Chen, Yuanhong and Liu, Fengbei and Wang, Hu and Wang, Chong and Liu, Yuyuan and Tian, Yu and Carneiro, Gustavo , title =. 2023 , pages =

    2023

  35. [35]

    Multi-Label Iterated Learning for Image Classification With Label Ambiguity , booktitle = CVPR, month =

    Rajeswar, Sai and Rodr. Multi-Label Iterated Learning for Image Classification With Label Ambiguity , booktitle = CVPR, month =. 2022 , pages =

    2022

  36. [36]

    2022 , pages =

    Ben-Baruch, Emanuel and Ridnik, Tal and Friedman, Itamar and Ben-Cohen, Avi and Zamir, Nadav and Noy, Asaf and Zelnik-Manor, Lihi , title =. 2022 , pages =

    2022

  37. [37]

    2022 , pages =

    Kim, Youngwook and Kim, Jae Myung and Akata, Zeynep and Lee, Jungwoo , title =. 2022 , pages =

    2022

  38. [38]

    2022 , pages =

    Zhang, Shu and Xu, Ran and Xiong, Caiming and Ramaiah, Chetan , title =. 2022 , pages =

    2022

  39. [39]

    2022 , pages =

    Liu, Ruyang and Liu, Hao and Li, Ge and Hou, Haodi and Yu, TingHao and Yang, Tao , title =. 2022 , pages =

    2022

  40. [40]

    2021 , pages =

    Zhao, Jiawei and Yan, Ke and Zhao, Yifan and Guo, Xiaowei and Huang, Feiyue and Li, Jia , title =. 2021 , pages =

    2021

  41. [41]

    2021 , pages =

    Ridnik, Tal and Ben-Baruch, Emanuel and Zamir, Nadav and Noy, Asaf and Friedman, Itamar and Protter, Matan and Zelnik-Manor, Lihi , title =. 2021 , pages =

    2021

  42. [42]

    2021 , pages =

    Lanchantin, Jack and Wang, Tianlu and Ordonez, Vicente and Qi, Yanjun , title =. 2021 , pages =

    2021

  43. [43]

    2023 , pages =

    Xu, Mengde and Zhang, Zheng and Wei, Fangyun and Hu, Han and Bai, Xiang , title =. 2023 , pages =

    2023

  44. [44]

    2023 , pages =

    Liang, Feng and Wu, Bichen and Dai, Xiaoliang and Li, Kunpeng and Zhao, Yinan and Zhang, Hang and Zhang, Peizhao and Vajda, Peter and Marculescu, Diana , title =. 2023 , pages =

    2023

  45. [45]

    2021 , eprint =

    CLIP-Adapter: Better Vision-Language Models with Feature Adapters , author =. 2021 , eprint =

    2021

  46. [46]

    2022 , pages =

    Zhou, Kaiyang and Yang, Jingkang and Loy, Chen Change and Liu, Ziwei , title =. 2022 , pages =

    2022

  47. [47]

    DualCoOp: Fast Adaptation to Multi-Label Recognition with Limited Annotations , url =

    Sun, Ximeng and Hu, Ping and Saenko, Kate , booktitle = NIPS, editor =. DualCoOp: Fast Adaptation to Multi-Label Recognition with Limited Annotations , url =. 2022 , bdsk-url-1 =

    2022

  48. [48]

    Zero-shot Visual Relation Detection via Composite Visual Cues from Large Language Models , year =

    Li, Lin and Xiao, Jun and Chen, Guikun and Shao, Jian and Zhuang, Yueting and Chen, Long , booktitle = NIPS, publisher =. Zero-shot Visual Relation Detection via Composite Visual Cues from Large Language Models , year =

  49. [49]

    ICML , pages =

    Learning Transferable Visual Models From Natural Language Supervision , author =. ICML , pages =. 2021 , editor =

    2021

  50. [50]

    2021 , pages =

    Cole, Elijah and Mac Aodha, Oisin and Lorieul, Titouan and Perona, Pietro and Morris, Dan and Jojic, Nebojsa , title =. 2021 , pages =

    2021

  51. [51]

    2023 , pages =

    Abdelfattah, Rabab and Guo, Qing and Li, Xiaoguang and Wang, Xiaofeng and Wang, Song , title =. 2023 , pages =

    2023

  52. [52]

    2023 , publisher =

    Semantic Contrastive Bootstrapping for Single-Positive Multi-label Recognition , author =. 2023 , publisher =

    2023

  53. [53]

    Durand, Thibaut and Mehrasa, Nazanin and Mori, Greg , title =

  54. [54]

    Structured Semantic Transfer for Multi-Label Recognition with Partial Labels , author =

  55. [55]

    Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels , author =

  56. [56]

    G2NetPL: Generic Game-Theoretic Network for Partial-Label Image Classification , author =

  57. [57]

    Microsoft COCO: Common Objects in Context , booktitle = ECCV, year =

    Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll. Microsoft COCO: Common Objects in Context , booktitle = ECCV, year =

  58. [58]

    and Van Gool, L

    Everingham, M. and Van Gool, L. and Williams, C. K. I. and Winn, J. and Zisserman, A. , title =

  59. [59]

    2011 , institution =

    The Caltech-UCSD Birds-200-2011 Dataset , author =. 2011 , institution =

    2011

  60. [60]

    NUS-WIDE: A Real-World Web Image Database from National University of Singapore , address =

    Tat-Seng Chua and Jinhui Tang and Richang Hong and Haojie Li and Zhiping Luo and Yan-Tao Zheng , booktitle =. NUS-WIDE: A Real-World Web Image Database from National University of Singapore , address =

  61. [61]

    and Cogswell, Michael and Das, Abhishek and Vedantam, Ramakrishna and Parikh, Devi and Batra, Dhruv , title =

    Selvaraju, Ramprasaath R. and Cogswell, Michael and Das, Abhishek and Vedantam, Ramakrishna and Parikh, Devi and Batra, Dhruv , title =

  62. [62]

    2022 , pages =

    Du, Yu and Wei, Fangyun and Zhang, Zihe and Shi, Miaojing and Gao, Yue and Li, Guoqi , title =. 2022 , pages =

    2022

  63. [63]

    Open-Vocabulary Multi-Label Classification via Multi-Modal Knowledge Transfer , volume =

    He, Sunan and Guo, Taian and Dai, Tao and Qiao, Ruizhi and Shu, Xiujun and Ren, Bo and Xia, Shu-Tao , year =. Open-Vocabulary Multi-Label Classification via Multi-Modal Knowledge Transfer , volume =. doi:10.1609/aaai.v37i1.25159 , abstractnote =

    work page doi:10.1609/aaai.v37i1.25159

  64. [64]

    PromptDet: Towards Open-Vocabulary Detection Using Uncurated Images , booktitle = ECCV, year =

    Feng, Chengjian and Zhong, Yujie and Jie, Zequn and Chu, Xiangxiang and Ren, Haibing and Wei, Xiaolin and Xie, Weidi and Ma, Lin , editor =. PromptDet: Towards Open-Vocabulary Detection Using Uncurated Images , booktitle = ECCV, year =

  65. [65]

    2023 , pages =

    Shi, Cheng and Yang, Sibei , title =. 2023 , pages =

    2023

  66. [66]

    arXiv preprint arXiv:2001.07685 , year=

    FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence , author=. arXiv preprint arXiv:2001.07685 , year=

    work page arXiv 2001

  67. [67]

    FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling , url =

    Zhang, Bowen and Wang, Yidong and Hou, Wenxin and WU, HAO and Wang, Jindong and Okumura, Manabu and Shinozaki, Takahiro , booktitle = NIPS, editor =. FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling , url =

  68. [68]

    Explanatory Object Part Aggregation for Zero-Shot Learning , year=

    Chen, Xin and Deng, Xiaoling and Lan, Yubin and Long, Yongbing and Weng, Jian and Liu, Zhiquan and Tian, Qi , journal=. Explanatory Object Part Aggregation for Zero-Shot Learning , year=

  69. [69]

    Semi-Supervised and Unsupervised Deep Visual Learning: A Survey , year=

    Chen, Yanbei and Mancini, Massimiliano and Zhu, Xiatian and Akata, Zeynep , journal=. Semi-Supervised and Unsupervised Deep Visual Learning: A Survey , year=

  70. [70]

    Classifier-guided CLIP Distillation for Unsupervised Multi-label Classification , author=

  71. [71]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Tagclip: A local-to-global framework to enhance open-vocabulary multi-label classification of clip without training , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  72. [72]

    BAC-GCN: Background-Aware CLIP-GCN Framework for Unsupervised Multi-Label Classification , author=

  73. [73]

    International conference on machine learning , pages=

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  74. [74]

    Semi-Supervised Classification with Graph Convolutional Networks

    Semi-supervised classification with graph convolutional networks , author=. arXiv preprint arXiv:1609.02907 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  75. [75]

    Ilya Loshchilov and Frank Hutter , title =

  76. [76]

    MambaML: Exploring State Space Models for Multi-Label Image Classification , author=

  77. [77]

    Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning , author=

  78. [78]

    Towards Space and Semantics: Object-Purified Representation Learning for Multi-Label Image Classification , author=

  79. [79]

    Recover and Match: Open-Vocabulary Multi-Label Recognition through Knowledge-Constrained Optimal Transport , author=