Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations

T0 review · 3 major / 5 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 ReTA outperforms prior cache-based test-time adaptation methods by reweighting entropy with a consistency vote and by reshaping decision boundaries as evolving Gaussians, achieving the best reported average accuracy on both benchmarks.

desk verdict Solid integrated TTA method; the real caveat is the per-dataset validation tuning behind the SOTA claim, not the consistency score. read the letter →

arxiv 2507.09500 v2 pith:RQVAO6EB submitted 2025-07-13 cs.CV

Yiwen Liang , Hui Chen , Yizhe Xiong , Zihan Zhou , Mengyao Lyu , Zijia Lin , Shuaicheng Niu , Sicheng Zhao
show 2 more authors
Jungong Han Guiguang Ding
This is my paper
classification cs.CV
keywords test-timeadaptationvision-languagemodelsCLIPcache-basedentropyreweightingpredictionconsistencydistributioncalibrationout-of-distributiongeneralization
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 claims that cache-based test-time adaptation of vision-language models fails for two reasons: entropy-based selection admits confidently wrong samples into the cache, and fixed text-embedding decision boundaries cannot accommodate large visual shifts. ReTA attacks both problems, using a stability-consistency score computed by a vote across adjacent class-specific text embeddings to reweight entropy during cache updates, and modeling each class's text embeddings as a multivariate Gaussian that evolves through residual learning to calibrate decision boundaries. On the standard benchmarks, ReTA reports average top-1 accuracy of 71.37% across ten cross-dataset tasks and 67.29% on ImageNet plus its four shifted variants with CLIP-ViT-B/16, ahead of the previous best cache-based methods. If the results hold, consistency voting over neighboring text embeddings, rather than raw confidence, is a reliable signal for keeping test-time caches clean.

What carries the argument

The central object is the stability-consistency score w (Eq. 13), which reweights prediction entropy during cache updates. To build it, the method sorts the hand-crafted prompts of each class by cumulative intra-class cosine similarity, progressively bins them into M adjacent class-specific text embeddings, projects image features into the text subspace via SVD to reduce the modality gap, and treats the M embeddings as a semantic voting committee whose pseudo-labels yield the score; w = 1 + log(RS) is minimal when all committee members agree and the majority vote matches the original prediction, and grows with disagreement or instability. The complementary mechanism, Diversity-driven Distribution Calibration (DDC), models each class's adjacent text embeddings as an approximate multivariate Gaussian N(μ, Σ), updates them through residual learning and progressive averaging, and uses the Gaussian means as a refined classifier with a surrogate loss that exploits the covariance terms. The score gates what enters the cache and what is allowed to update the residuals, while DDC reshapes the decision boundary.

What would settle it

Run ReTA on ImageNet-A, where CLIP is overconfident on many natural adversarial images, and measure the accuracy of the samples admitted to the cache with w = 1; if that cache accuracy is no better than chance while ReTA still improves, the gains come from the Gaussian boundary rather than the consistency signal. A cleaner experiment is a synthetic shift engineered so CLIP is consistently confident and wrong — for instance heavy blur combined with color negation — and check whether ReTA then falls below an entropy-only cache baseline.

Watch

Extended reading notes

Core claim

ReTA's central claim is that both failure modes of cache-based test-time adaptation can be addressed without labels or backbone updates. For the cache, instead of ranking samples by raw prediction entropy, ReTA computes a stability-consistency score w = 1 + log(RS) from agreement among pseudo-labels produced by M adjacent class-specific text embeddings and agreement of the majority vote with the original prediction; this score multiplies the entropy, so unstable or inconsistent samples are preferentially evicted and only samples with w = 1 drive residual updates. For the decision boundary, the adjacent text embeddings of each class are treated as an approximate multivariate Gaussian, whose means are evolved with learnable residuals and progressive averaging, and the resulting Gaussian logits are combined with cache logits and CLIP logits for the final prediction. The paper reports that this combination consistently outperforms state-of-the-art cache-based methods such as DPE and BoostAdapter on both cross-dataset generalization and natural distribution shift benchmarks.

Load-bearing premise

The load-bearing premise is that the stability-consistency score w really marks which test samples carry correct pseudo-labels, because both the cache and the Gaussian calibration are updated only from samples the score trusts; if agreement across the adjacent text embeddings stops correlating with correctness under a particular shift, the method's own training signal becomes polluted.

Editorial extensions

If this is right

  • On Cross-Datasets with CLIP-ViT-B/16, ReTA reports 71.37% average accuracy versus 69.40% for DPE, the previous best cache-based method, and 67.29% versus 66.19% for BoostAdapter on ImageNet and its four shifted variants.
  • Cache quality stays high throughout testing under ReTA, while earlier methods' caches fill with misclassified low-entropy samples as adaptation proceeds (Figure 1 and the qualitative cache comparisons in Figure 7).
  • The Gaussian decision boundary carries more of the gain than cache refinement: ablations attribute larger improvements to DDC than to CER, and removing the Gaussian logits flips 6.18% of predictions.
  • The gains are not specific to CLIP-ViT-B/16: ReTA also improves over DPE and TDA on EVA-02-B-16 and CoCa-ViT-B-32, and it beats RLCF when the entropy objective is swapped for CLIP reward feedback, with a lower average Expected Calibration Error than TDA and DPE.

Reading between the lines

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

  • The same consistency-voting signal could serve as a pseudo-label filter in other self-training and source-free adaptation settings, where confident-but-wrong samples cause the same error accumulation that ReTA targets.
  • Because DDC contributes more than CER in the ablations, a plausible reading is that the residually updated text embeddings carry most of the adaptation; a frozen-cache variant of ReTA would isolate how much of the gain is due to the cache alone.
  • The committee is built from a fixed set of hand-crafted prompts plus CuPL, so the score's reliability is coupled to prompt quality; perturbing or shrinking that prompt set is a direct stress test of the consistency assumption.
  • Since per-sample cost is dominated by 64 augmented views and the SVD projection, a batched or adaptive-view variant of the consistency assessment could cut inference time while preserving the reliability signal.
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 / 5 minor

Summary. The paper proposes ReTA, a cache-based test-time adaptation method for CLIP-style vision-language models. The method has two components: Consistency-aware Entropy Reweighting (CER), which uses a stability-consistency score across adjacent text embeddings to reweight entropy when updating a test-time cache; and Diversity-driven Distribution Calibration (DDC), which models class-wise text embeddings as multivariate Gaussians with residual learning to obtain more flexible decision boundaries. The final prediction combines cache logits with Gaussian-calibrated logits. The authors evaluate on Cross-Datasets generalization and natural distribution shifts (ImageNet and four variants) with CLIP-RN50 and CLIP-ViT-B/16, and report state-of-the-art averages (e.g., 71.37% on Cross-Datasets with ViT-B/16 versus 69.40% for DPE, and 67.29% on natural shifts versus 66.19% for BoostAdapter). They also report ablations, ECE, runtime, and experiments with EVA-02 and CoCa.

Significance. If the results hold, ReTA is a practically useful contribution to cache-based TTA: it avoids encoder backpropagation, improves cache quality under distribution shift, and provides better calibration than prior cache-based methods. The paper's strengths include a broad evaluation across two backbones and two additional model families, a public code release, runtime comparisons, ECE measurements, and careful ablations of the proposed components. The main caveat is that the headline SOTA claim currently rests on an evaluation-protocol issue: Appendix B discloses per-dataset selection of alpha, beta, and eta using validation accuracy, which conflicts with the unlabeled TTA setting and makes the comparison with fixed-parameter baselines not apples-to-apples. The consistency-correctness assumption is a further risk, though the ablation evidence suggests it is not the sole driver of the gains.

major comments (3)
  1. [Appendix B] Appendix B states that the affine cache parameters alpha and beta are dataset-specific and that the balance weight eta is selected by 'maximizing the average validation accuracy.' This is a load-bearing protocol issue: the paper advertises unlabeled test-time adaptation, but a label-guided search over at least three hyperparameters per dataset can plausibly account for the small reported margins (1.97 pp over DPE on Cross-Datasets and 1.10 pp over BoostAdapter on natural shifts with ViT-B/16). The authors should state clearly whether the validation split is labeled, report results with a single fixed set of hyperparameters across all datasets, or, if tuning is retained, demonstrate that the method is insensitive to the choice and that all baselines were tuned under the same protocol.
  2. [Section 4.1] The paper states that all experiments were run with three random seeds and that the average is reported, but no standard deviations or per-seed results appear in Tables 1-6. Since the SOTA margins are only one to two percentage points, the absence of variance estimates leaves it unclear whether the ranking over DPE and BoostAdapter is statistically stable. Please report per-seed results or at least standard deviations for the main tables.
  3. [Section 3.2 and Algorithm 1] The stability-consistency score w (Eq. 13) gates both cache prioritization (Eq. 14) and the residual updates in DDC (Algorithm 1, line 10). The method therefore assumes that prediction consistency across adjacent text embeddings is positively correlated with pseudo-label correctness under distribution shift. The supporting evidence is indirect: aggregate cache-accuracy curves (Figure 1), a few qualitative cache examples (Figure 7), and ablations of the score formulation (Tables 8-10). Please add a quantitative analysis of precision at w=1 versus w>1, ideally with a per-dataset breakdown on ImageNet-A and other hard OOD sets, to show that confidently wrong samples do not contaminate the cache or the DDC updates.
minor comments (5)
  1. [References] References [3] and [4] are the same work, and references [26], [27], and [28] are also the same work; please deduplicate them.
  2. [Appendix C] Tables 8 and 9 are identical; one of them appears to be a duplicate and should be removed.
  3. [Tables 2 and 7] The BoostAdapter average accuracy is reported as 66.19 in Table 2 but 66.34 in Table 7; please reconcile these numbers and state which setup each refers to.
  4. [Figure 1] The caption does not specify the dataset, backbone, or number of test samples for the cache-accuracy curves shown in Figure 1; please provide this information so the reader can judge how representative the example is.
  5. [Algorithm 1] In Algorithm 1, line 7, the condition max(H'(Fcache)) is ambiguous until the cache is full; please specify whether the comparison is made per class and how the first SZ samples per class are admitted.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ReTA's headline results are measured against external labeled benchmarks, and its CER and DDC components are evaluated empirically rather than derived from their own definitions.

full rationale

The claimed derivation chain is not circular. ReTA's central claim—higher top-1 accuracy on Cross-Datasets and natural-shift benchmarks—is evaluated against external labeled test sets (Tables 1 and 2), so the headline numbers are not recoverable by construction from the proposed equations. CER's consistency score w (Eq. 13) is a sample-selection heuristic computed from adjacent text embeddings, and the cache is populated by that heuristic; the quality of the choice is judged by downstream accuracy on real labels, not by the heuristic itself. DDC's Gaussian calibration (Eqs. 16–23) is an optimization procedure whose success is measured against held-out test accuracy; no equation defines the reported accuracy in terms of the fitted residuals or balance weights. The one flagged in-scope weakness is Appendix B, which admits dataset-specific α, β, and η selected by maximizing average validation accuracy; this is a label-use/protocol concern rather than a circular reduction and does not by itself make the benchmark result equivalent to an input. The self-training aspect—samples with w=1 update the same text embeddings used to compute w—is standard test-time entropy/consistency optimization and does not make the external benchmark result definitionally forced. Self-citations appear only as background or as links to external method components (DPE, ProDA, TDA), and no load-bearing uniqueness theorem or ansatz is imported solely from the authors' own prior work. Therefore the paper is self-contained against external benchmarks and the circularity score is 0.

Assumptions & free parameters 9 free parameters · 6 assumptions · 0 invented entities

The method introduces no new entities; it postulates a reliability signal (consistency across prompt neighborhoods) and a Gaussian model of text embeddings, both built on assumptions about CLIP embedding geometry. The central claim depends on the ad hoc assumption that consistency implies correctness, which is the method's main risk.

free parameters (9)
  • alpha and beta (cache logit modulation) = dataset-specific, inherited from TDA/DPE, not listed in the paper
    These control the affine function A(x)=alpha*exp(-beta*(1-x)) in Eq. 4; different values per dataset are used, so the gains partly depend on dataset-specific tuning.
  • eta (final prediction balance weight) = optimal in [0.2, 0.6], chosen per benchmark
    In Eq. 23, eta balances cache logits and Gaussian logits; selected to maximize average validation accuracy (Appendix B).
  • gamma (consistency penalty) = 2
    Penalty applied when majority-vote label mismatches the original prediction (Eq. 11); value chosen by ablation (Table 10).
  • delta (entropy threshold for augmented views) = 0.1
    Filters augmented views used in L_ent (Eq. 19); fixed, from TPT-style settings.
  • M (number of adjacent text embeddings) = 3
    Size of semantic voting committee and Gaussian components; chosen by hyperparameter analysis (Figure 5b).
  • n (number of retained SVD components) = 64
    Dimensionality of the text subspace projection (Eq. 7); chosen by hyperparameter analysis (Figure 5a).
  • cache size = 3 samples per class
    Capacity of the per-class cache; chosen by ablation (Figure 6a).
  • lambda1, lambda2 (loss weights) = 0.3, 0.02
    Weights of surrogate and alignment losses in Eq. 18; tuned via sensitivity analysis (Figure 5c).
  • tau_c (confidence threshold for global updates) = not specified in the text
    Samples with entropy below this threshold update the global text prototype (Eq. 17); the threshold is fixed but its value is not reported.
assumptions (6)
  • domain assumption Class-wise text embedding neighborhoods, built from multiple hand-crafted prompts, are semantically meaningful for assessing prediction reliability.
    CER constructs adjacent embeddings by progressive binning of prompt embeddings (Sec 3.2); this assumes prompt paraphrases form a useful local structure.
  • ad hoc to paper Prediction consistency across adjacent text embeddings is positively correlated with pseudo-label correctness under distribution shift.
    Core premise of Eq. 13-14 and Algorithm 1; samples with w=1 are trusted for cache and DDC training, but the paper only validates this indirectly (Figures 1 and 7).
  • domain assumption The set of adjacent text embeddings can be approximated by a multivariate Gaussian distribution per class.
    DDC inherits the Gaussian prompt assumption from ProDA [33]; with M=3 prompts per class, the covariance estimate is extremely coarse.
  • domain assumption Projecting image features into the top-64 SVD text subspace preserves the semantic structure needed for similarity-based voting.
    Phi_proj = V^T V in Eq. 7 is applied to image features only; this follows [68] and assumes the text subspace is a valid reference for images despite the modality gap.
  • domain assumption AugMix-augmented views are label-preserving and their entropy can serve as a safe training signal for text residuals.
    L_ent (Eq. 19) is computed over 63 augmented views, following TPT [46]; this assumes augmentations do not change semantics.
  • domain assumption The confidence thresholds (delta, tau_c) and gating by w=1 prevent degradation of the self-training loop.
    The method relies on these filters to avoid error accumulation in entropy minimization; no analysis of the self-training dynamics is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations." pith.science (2026). https://pith.science/paper/RQVAO6EB

@misc{pith2026250709500,
  author       = {Pith},
  title        = {Pith review of: Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RQVAO6EB}},
  note         = {Machine review of arXiv:2507.09500}
}
read the original abstract

Vision-language models (VLMs) exhibit remarkable zero-shot capabilities but struggle with distribution shifts in downstream tasks when labeled data is unavailable, which has motivated the development of Test-Time Adaptation (TTA) to improve VLMs' performance during inference without annotations. Among various TTA approaches, cache-based methods show promise by preserving historical knowledge from low-entropy samples in a dynamic cache and fostering efficient adaptation. However, these methods face two critical reliability challenges: (1) entropy often becomes unreliable under distribution shifts, causing error accumulation in the cache and degradation in adaptation performance; (2) the final predictions may be unreliable due to inflexible decision boundaries that fail to accommodate large downstream shifts. To address these challenges, we propose a Reliable Test-time Adaptation (ReTA) method that integrates two complementary strategies to enhance reliability from two perspectives. First, to mitigate the unreliability of entropy as a sample selection criterion for cache construction, we introduce Consistency-aware Entropy Reweighting (CER), which incorporates consistency constraints to weight entropy during cache updating. While conventional approaches rely solely on low entropy for cache prioritization and risk introducing noise, our method leverages predictive consistency to maintain a high-quality cache and facilitate more robust adaptation. Second, we present Diversity-driven Distribution Calibration (DDC), which models class-wise text embeddings as multivariate Gaussian distributions, enabling adaptive decision boundaries for more accurate predictions across visually diverse content. Extensive experiments demonstrate that ReTA consistently outperforms state-of-the-art methods, particularly under real-world distribution shifts. Code: https://github.com/Evelyn1ywliang/ReTA.

Figures

Figures reproduced from arXiv: 2507.09500 by the authors.

Figure 2
Figure 2. Decision boundary visualization. Left: Conventional [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overall framework of our proposed Reliable Test-time Adaptation (ReTA) approach under visual variations. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Final comparison of adaptation performance and [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Hyperparameter analysis. (a) Number of singular [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Sensitivity analysis of hyper-parameters. (a) Different cache size. (b) Different number of augmented views. (c) Effect [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Visual comparison of cached samples with their corresponding entropy values (shown below each image). Samples [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Textual prompts examples in our experiments. In [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Multi-Cache Enhanced Prototype Learning for Test-Time Generalization of Vision-Language Models

    cs.CV 2025-08 unverdicted novelty 5.0 of 10

    The submitted full text does not match the abstract, so the manuscript cannot be assessed as a coherent preprint.

Reference graph

Works this paper leans on

73 extracted references · 59 canonical work pages · cited by 1 Pith paper

  1. [1]

    Jameel Abdul Samadh, Mohammad Hanan Gani, Noor Hussein, Muhammad Uzair Khattak, Muhammad Muzammal Naseer, Fahad Shahbaz Khan, and Salman H Khan. 2023. Align Your Prompts: Test-Time Prompting with Distribution Align- ment for Zero-Shot Generalization. In Advances in Neural Information Processing Systems, A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt...

    work page 2023

  2. [2]

    Sandhini Agarwal, Gretchen Krueger, Jack Clark, Alec Radford, Jong Wook Kim, and Miles Brundage. 2021. Evaluating clip: towards characterization of broader capabilities and downstream implications. arXiv preprint arXiv:2108.02818 (2021)

    arXiv 2021

  3. [3]

    Wentao Bao, Lichang Chen, Heng Huang, and Yu Kong. 2025. Prompting Language-Informed Distribution for Compositional Zero-Shot Learning. In Com- puter Vision – ECCV 2024 . Cham, 107–123

    work page 2025

  4. [4]

    Wentao Bao, Lichang Chen, Heng Huang, and Yu Kong. 2025. Prompting Language-Informed Distribution for Compositional Zero-Shot Learning. In Com- puter Vision – ECCV 2024 . Springer Nature Switzerland, Cham, 107–123

    work page 2025

  5. [5]

    Lukas Bossard, Matthieu Guillaumin, and Luc Van Gool. 2014. Food-101 – Min- ing Discriminative Components with Random Forests. In Computer Vision – ECCV 2014, David Fleet, Tomas Pajdla, Bernt Schiele, and Tinne Tuytelaars (Eds.). Springer International Publishing, Cham, 446–461

    work page 2014

  6. [6]

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, and Jenia Jitsev

  7. [7]

    Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos, Sammy Mohamed, and An- drea Vedaldi. 2014. Describing Textures in the Wild. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    work page 2014

  8. [8]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Im- ageNet: A large-scale hierarchical image database. In 2009 IEEE Conference on Computer Vision and Pattern Recognition . 248–255

    work page 2009

Show all 73 references
  1. [9]

    Zixuan Ding, Ao Wang, Hui Chen, Qiang Zhang, Pengzhang Liu, Yongjun Bao, Weipeng Yan, and Jungong Han. 2023. Exploring Structured Semantic Prior for Multi Label Recognition With Incomplete Labels. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni...

    2023

  2. [10]

    Yuxin Fang, Quan Sun, Xinggang Wang, Tiejun Huang, Xinlong Wang, and Yue Cao. 2024. EVA-02: A visual representation for neon genesis. Image and Vision Computing 149 (2024), 105171. doi:10.1016/j.imavis.2024.105171

    2024

  3. [11]

    Matteo Farina, Gianni Franchi, Giovanni Iacca, Massimiliano Mancini, and Elisa Ricci. [n. d.]. Frustratingly Easy Test-Time Adaptation of Vision-Language Models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems

  4. [12]

    Fergus, and P

    Li Fei-Fei, R. Fergus, and P. Perona. 2004. Learning Generative Visual Models from Few Training Examples: An Incremental Bayesian Approach Tested on 101 Object Categories. In 2004 Conference on Computer Vision and Pattern Recognition Workshop. 178–178

    2004

  5. [13]

    Chun-Mei Feng, Kai Yu, Yong Liu, Salman Khan, and Wangmeng Zuo. 2023. Di- verse Data Augmentation with Diffusions for Effective Test-time Prompt Tuning. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 2704–2714

    2023

  6. [14]

    Constanza Fierro and Anders Søgaard. 2022. Factual Consistency of Multilingual Pretrained Language Models. In Findings of the Association for Computational Linguistics: ACL 2022. Association for Computational Linguistics, Dublin, Ireland, 3046–3052. doi:10.18653/v1/2022.findin...

    2022 doi

  7. [15]

    Yunhao Ge, Jie Ren, Andrew Gallagher, Yuxiao Wang, Ming-Hsuan Yang, Hartwig Adam, Laurent Itti, Balaji Lakshminarayanan, and Jiaping Zhao. 2023. Improving Zero-Shot Generalization and Robustness of Multi-Modal Models. In Proceedings of the IEEE/CVF Conference on Computer Visio...

    2023

  8. [16]

    Tianxiang Hao, Hui Chen, Yuchen Guo, and Guiguang Ding. 2023. Consol- idator: Mergeable Adapter with Grouped Connections for Visual Adaptation. arXiv:2305.00603 [cs.CV] https://arxiv.org/abs/2305.00603

    2023 arXiv

  9. [17]

    Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. 2019. EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 12, 7 (2019), 2217–2226

    2019

  10. [18]

    Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kadavath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu, Samyak Parajuli, Mike Guo, Dawn Song, Jacob Steinhardt, and Justin Gilmer. 2021. The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization....

    2021

  11. [19]

    Dan Hendrycks*, Norman Mu*, Ekin Dogus Cubuk, Barret Zoph, Justin Gilmer, and Balaji Lakshminarayanan. 2020. AugMix: A Simple Method to Improve Robustness and Uncertainty under Data Shift. In International Conference on Learning Representations

    2020

  12. [20]

    Dan Hendrycks, Kevin Zhao, Steven Basart, Jacob Steinhardt, and Dawn Song

  13. [21]

    Tony Huang, Jack Chu, and Fangyun Wei. 2022. Unsupervised Prompt Learning for Vision-Language Models. arXiv:2204.03649 https://arxiv.org/abs/2204.03649

    2022 arXiv

  14. [22]

    Ahmet Iscen, Mathilde Caron, Alireza Fathi, and Cordelia Schmid. 2024. Retrieval- Enhanced Contrastive Vision-Text Models. InThe Twelfth International Conference on Learning Representations

    2024

  15. [23]

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc Le, Yun-Hsuan Sung, Zhen Li, and Tom Duerig. 2021. Scaling Up Visual and Vision- Language Representation Learning With Noisy Text Supervision. In Proceedings of the 38th International Conference on Mac...

    2021

  16. [24]

    Jiahaoli, Yang Lu, Yuan Xie, and Yanyun Qu. 2024. Relationship Prompt Learning is Enough for Open-Vocabulary Semantic Segmentation. In The Thirty-eighth Annual Conference on Neural Information Processing Systems

    2024

  17. [25]

    Adilbek Karmanov, Dayan Guan, Shijian Lu, Abdulmotaleb El Saddik, and Eric Xing. 2024. Efficient Test-Time Adaptation of Vision-Language Models. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 14162–14171

    2024

  18. [28]

    Zaid Khan and Yun Fu. 2024. Consistency and Uncertainty: Identifying Unreliable Responses From Black-Box Vision-Language Models for Selective Visual Question Answering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 10854–10863

    2024

  19. [29]

    Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 2013. 3D Object Repre- sentations for Fine-Grained Categorization. In Proceedings of the IEEE Interna- tional Conference on Computer Vision (ICCV) Workshops

    2013

  20. [30]

    Victor Weixin Liang, Yuhui Zhang, Yongchan Kwon, Serena Yeung, and James Y Zou. 2022. Mind the Gap: Understanding the Modality Gap in Multi-modal Con- trastive Representation Learning. In Advances in Neural Information Processing Systems, S. Koyejo, S. Mohamed, A. Agarwal, D. ...

    2022

  21. [31]

    Chang Liu, Lichen Wang, and Yun Fu. 2023. Rethinking Neighborhood Consis- tency Learning on Unsupervised Domain Adaptation. In Proceedings of the 31st ACM International Conference on Multimedia (Ottawa ON, Canada) (MM ’23). Association for Computing Machinery, New York, NY, US...

    2023

  22. [32]

    Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    2017 arXiv

  23. [33]

    Yuning Lu, Jianzhuang Liu, Yonggang Zhang, Yajing Liu, and Xinmei Tian. 2022. Prompt Distribution Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 5206–5215

    2022

  24. [34]

    Mengyao Lyu, Tianxiang Hao, Xinhao Xu, Hui Chen, Zijia Lin, Jungong Han, and Guiguang Ding. 2025. Learn from the Learnt: Source-Free Active Domain Adaptation via Contrastive Sampling and Visual Persistence. In Computer Vision – ECCV 2024. Springer Nature Switzerland, Cham, 228–246

    2025

  25. [35]

    Mengyao Lyu, Jundong Zhou, Hui Chen, Yijie Huang, Dongdong Yu, Yaqian Li, Yandong Guo, Yuchen Guo, Liuyu Xiang, and Guiguang Ding. 2023. Box-Level Active Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 23766–23775

    2023

  26. [36]

    XIAOSONG MA, Jie ZHANG, Song Guo, and Wenchao Xu. 2023. SwapPrompt: Test-Time Prompt Adaptation for Vision-Language Models. InAdvances in Neural Information Processing Systems , A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine (Eds.), Vol. 36. Curran Associa...

    2023

  27. [37]

    Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew Blaschko, and Andrea Vedaldi

  28. [38]

    Maria-Elena Nilsback and Andrew Zisserman. 2008. Automated Flower Clas- sification over a Large Number of Classes. In 2008 Sixth Indian Conference on Computer Vision, Graphics & Image Processing . 722–729

    2008

  29. [39]

    Letitia Parcalabescu and Anette Frank. 2025. Do Vision & Language Decoders use Images and Text equally? How Self-consistent are their Explanations?. In The Thirteenth International Conference on Learning Representations . https: //openreview.net/forum?id=lCasyP21Bf

    2025

  30. [40]

    Omkar M Parkhi, Andrea Vedaldi, Andrew Zisserman, and C. V. Jawahar. 2012. Cats and dogs. In2012 IEEE Conference on Computer Vision and Pattern Recognition. 3498–3505. MM ’25, October 27–31, 2025, Dublin, Ireland Yiwen Liang et al

    2012

  31. [41]

    Sarah Pratt, Ian Covert, Rosanne Liu, and Ali Farhadi. 2023. What Does a Platypus Look Like? Generating Customized Prompts for Zero-Shot Image Classification. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 15691–15701

    2023

  32. [42]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. In Proceedings...

    2021

  33. [43]

    Benjamin Recht, Rebecca Roelofs, Ludwig Schmidt, and Vaishaal Shankar. 2019. Do ImageNet Classifiers Generalize to ImageNet?. In Proceedings of the 36th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 97), Kamalika Chaudhuri and Rus...

    2019

  34. [44]

    H. S. Seung, M. Opper, and H. Sompolinsky. 1992. Query by committee. InProceed- ings of the Fifth Annual Workshop on Computational Learning Theory (Pittsburgh, Pennsylvania, USA) (COLT ’92). Association for Computing Machinery, New York, NY, USA, 287–294

    1992

  35. [45]

    Welle, Mårten Björkman, and Danica Kragic

    Peiyang Shi, Michael C. Welle, Mårten Björkman, and Danica Kragic. 2023. To- wards understanding the modality gap in CLIP. In ICLR 2023 Workshop on Multi- modal Representation Learning: Perks and Pitfalls

    2023

  36. [46]

    Manli Shu, Weili Nie, De-An Huang, Zhiding Yu, Tom Goldstein, Anima Anand- kumar, and Chaowei Xiao. 2022. Test-Time Prompt Tuning for Zero-Shot General- ization in Vision-Language Models. In Advances in Neural Information Processing Systems, S. Koyejo, S. Mohamed, A. Agarwal, ...

    2022

  37. [47]

    Khurram Soomro, Amir Roshan Zamir, and Mubarak Shah. 2012. A dataset of 101 human action classes from videos in the wild. Center for Research in Computer Vision 2, 11 (2012), 1–7

    2012

  38. [48]

    Elaine Sui, Xiaohan Wang, and Serena Yeung-Levy. 2025. Just Shift It: Test-Time Prototype Shifting for Zero-Shot Generalization with Vision-Language Models. In Proceedings of the Winter Conference on Applications of Computer Vision (W ACV) . 825–835

    2025

  39. [49]

    Ao Wang, Hui Chen, Zijia Lin, Zixuan Ding, Pengzhang Liu, Yongjun Bao, Weipeng Yan, and Guiguang Ding. 2023. Hierarchical Prompt Learning Us- ing CLIP for Multi-label Classification with Single Positive Labels. In Proceedings of the 31st ACM International Conference on Multime...

    2023

  40. [50]

    Ao Wang, Hui Chen, Zijia Lin, Jungong Han, and Guiguang Ding. 2025. LSNet: See Large, Focus Small. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 9718–9729

    2025

  41. [51]

    Haohan Wang, Songwei Ge, Zachary Lipton, and Eric P Xing. 2019. Learning Ro- bust Global Representations by Penalizing Local Predictive Power. InAdvances in Neural Information Processing Systems, H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett...

    2019

  42. [52]

    Handing Wang, Yaochu Jin, and John Doherty. 2017. Committee-Based Ac- tive Learning for Surrogate-Assisted Particle Swarm Optimization of Expensive Problems. IEEE Transactions on Cybernetics 47, 9 (2017), 2664–2677

    2017

  43. [53]

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-Consistency Improves Chain of Thought Reasoning in Language Models. arXiv:2203.11171 [cs.CL] https://arxiv.org/abs/2203.11171

    2023 arXiv

  44. [54]

    Ehinger, Aude Oliva, and Antonio Torralba

    Jianxiong Xiao, James Hays, Krista A. Ehinger, Aude Oliva, and Antonio Torralba

  45. [55]

    Chen-Wei Xie, Siyang Sun, Xiong Xiong, Yun Zheng, Deli Zhao, and Jingren Zhou. 2023. RA-CLIP: Retrieval Augmented Contrastive Language-Image Pre- Training. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 19265–19274

    2023

  46. [56]

    Yizhe Xiong, Hui Chen, Tianxiang Hao, Zijia Lin, Jungong Han, Yuesong Zhang, Guoxin Wang, Yongjun Bao, and Guiguang Ding. 2025. PYRA: Parallel Yielding Re-activation for Training-Inference Efficient Task Adaptation. In Computer Vision – ECCV 2024 . Springer Nature Switzerland,...

    2025

  47. [57]

    Yizhe Xiong, Hui Chen, Zijia Lin, Sicheng Zhao, and Guiguang Ding. 2023. Confidence-based Visual Dispersal for Few-shot Unsupervised Domain Adapta- tion. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 11621–11631

    2023

  48. [58]

    Jiahui Yu, Zirui Wang, Vijay Vasudevan, Legg Yeung, Mojtaba Seyedhosseini, and Yonghui Wu. 2022. CoCa: Contrastive Captioners are Image-Text Foundation Models. Transactions on Machine Learning Research (2022). https://openreview. net/forum?id=Ee277P3AYC

    2022

  49. [59]

    Tao Yu, Zhihe Lu, Xin Jin, Zhibo Chen, and Xinchao Wang. 2023. Task Residual for Tuning Vision-Language Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 10899–10909

    2023

  50. [60]

    Sycara, and Yaqi Xie

    Ce Zhang, Simon Stepputtis, Katia P. Sycara, and Yaqi Xie. 2024. Dual Prototype Evolving for Test-Time Generalization of Vision-Language Models. InThe Thirty- eighth Annual Conference on Neural Information Processing Systems

    2024

  51. [61]

    Renrui Zhang, Wei Zhang, Rongyao Fang, Peng Gao, Kunchang Li, Jifeng Dai, Yu Qiao, and Hongsheng Li. 2022. Tip-Adapter: Training-Free Adaption of CLIP for Few-Shot Classification. In Computer Vision – ECCV 2022 . Springer Nature Switzerland, Cham, 493–510

    2022

  52. [62]

    Taolin Zhang, Jinpeng Wang, Hang Guo, Tao Dai, Bin Chen, and Shu-Tao Xia

  53. [63]

    Yabin Zhang, Wenjie Zhu, Hui Tang, Zhiyuan Ma, Kaiyang Zhou, and Lei Zhang

  54. [64]

    Shuai Zhao, Xiaohan Wang, Linchao Zhu, and Yi Yang. 2024. Test-Time Adapta- tion with CLIP Reward for Zero-Shot Generalization in Vision-Language Models. In The Twelfth International Conference on Learning Representations

    2024

  55. [65]

    Kaiyang Zhou, Jingkang Yang, Chen Change Loy, and Ziwei Liu. 2022. Conditional Prompt Learning for Vision-Language Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 16816–16825

    2022

  56. [66]

    Kaiyang Zhou, Jingkang Yang, Chen Change Loy, and Ziwei Liu. 2022. Learning to prompt for vision-language models. International Journal of Computer Vision 130, 9 (2022), 2337–2348

    2022

  57. [67]

    Wenjie Zhou, Zhenxin Ding, Xiaodong Zhang, Haibo Shi, Junfeng Wang, and Dawei Yin. 2024. GOVERN: Gradient Orientation Vote Ensemble for Multi- Teacher Reinforced Distillation. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: Industry Tr...

    2024 doi

  58. [68]

    Xingyu Zhu, Beier Zhu, Yi Tan, Shuo Wang, Yanbin Hao, and Hanwang Zhang

  59. [69]

    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Dual Memory Networks: A Versatile Adaptation Approach for Vision- Language Models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 28718–28728

  60. [75]

    sewing machine

    Selective Vision-Language Subspace Projection for Few-shot CLIP. In Proceedings of the 32nd ACM International Conference on Multimedia (Melbourne VIC, Australia) (MM ’24). Association for Computing Machinery, New York, NY, USA, 3848–3857. Advancing Reliable Test-Time Adaptatio...

    2025

  61. [2010]

    In 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition

    SUN database: Large-scale scene recognition from abbey to zoo. In 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition . 3485–3492

    2010

  62. [2013]

    arXiv:1306.5151 https: //arxiv.org/abs/1306.5151

    Fine-Grained Visual Classification of Aircraft. arXiv:1306.5151 https: //arxiv.org/abs/1306.5151

    arXiv

  63. [2021]

    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Natural Adversarial Examples. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 15262–15271

  64. [2023]

    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Reproducible Scaling Laws for Contrastive Language-Image Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2818–2829

  65. [2024]

    In The Thirty-eighth Annual Conference on Neural Infor- mation Processing Systems

    BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping. In The Thirty-eighth Annual Conference on Neural Infor- mation Processing Systems

Pith tools

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