Pith. sign in

REVIEW 3 major objections 5 minor 91 references

Respect Your Zero-Shot Uncertainty: Conservative Calibration for Test-Time-Adapted Vision-Language Models

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper shows that test-time adaptation sharpens predictions even when nothing else changes, and that restoring each sample's zero-shot entropy by minimal temperature scaling — ZAEC — lowers calibration error without altering accuracy.

desk verdict A simple, honest post-hoc calibration layer with a real diagnostic behind it, but the zero-shot entropy anchor has a documented failure mode that the paper underplays. read the letter →

arxiv 2608.05945 v1 pith:NOK35E26 submitted 2026-08-06 cs.CV

Jingyan Jiang , Yaru Sun , Xiao Chen , Jiazhen Huang , Caiting Li , Zhijian He , Yin Chen , Pingting Hao This is my paper
classification cs.CV
keywords test-timeadaptationvision-languagemodelscalibrationexpectederrorpredictiveentropytemperaturescalingzero-shotreferencedistributionshift
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

Test-time adaptation (TTA) makes vision-language models more accurate but often wrecks their calibration: confidence grows faster than correctness. The paper shows this happens even when the top-1 prediction is unchanged, a failure mode it calls prediction-preserving sharpening, and that the degree of sharpening relative to the paired zero-shot prediction tracks the rise in Expected Calibration Error. Building on that signal, it proposes ZAEC, a post-hoc method that, for each sample whose adapted entropy fell below its zero-shot entropy, raises the temperature just enough to restore that entropy, leaving all other predictions untouched. ZAEC needs no labels or learned parameters and provably preserves class rankings and accuracy. Across five TTA methods and 15 datasets, it reports the lowest macro-average post-hoc ECE on ViT-B/16, with consistent gains on RN50.

What carries the argument

The load-bearing object is a sample-specific entropy floor defined by the paired zero-shot prediction. The paper measures sharpening as the normalized entropy change δH_i = (H(p_TTA_i) − H(p_ZS_i))/log K, and for samples in $I^{{−}}$ = {i : H(p_TTA_i) < H(p_ZS_i)} solves for T*_i = min{T ≥ 1 : H(softmax(z_TTA_i/T)) = H(p_ZS_i)} by bisection, leaving samples outside $I^{{−}}$ at T = 1. Proposition 1 ensures temperature scaling strictly increases entropy and preserves the complete class ranking; Proposition 2 shows the chosen temperature is the unique minimum feasible one, is the closest feasible point to the raw TTA output in forward KL divergence among temperature-scaled predictions, and restores entropy exactly to max(H(p_TTA_i), H(p_ZS_i)). The mechanism therefore acts only on the concentration of the predictive distribution, never on its decisions.

What would settle it

Compute, for any held-out dataset and TTA method, the paired per-sample entropy changes and ECE: a single dataset–method pair with strongly negative mean relative entropy change and a decrease in ECE would refute the core association on which ZAEC is built.

Watch

Extended reading notes

Core claim

The paper's central claim is that test-time adaptation injects confidence that the underlying correctness does not support, and that the drop in predictive entropy relative to the paired zero-shot prediction is a label-free, sample-specific measurement of that unsupported sharpening. It establishes that, for samples whose top-1 prediction is unchanged by adaptation, confidence rises by 8.78–29.74 points while ECE rises by 8.59–20.34 points on ImageNet-A, and that across dataset–method pairs the mean entropy change and the ECE change are strongly correlated (r = −0.94). On that basis it defines ZAEC: for each sample whose adapted entropy fell below its zero-shot entropy, apply the smallest temperature T ≥ 1 that restores the zero-shot entropy; leave every other prediction untouched. The paper claims this one-sided, label-free correction achieves the lowest macro-average post-hoc ECE among compared post-hoc methods on ViT-B/16 across five TTA methods and 15 datasets, while preserving rankings, top-1 predictions, and accuracy, with consistent gains on RN50.

Load-bearing premise

The method assumes each sample's zero-shot entropy is the right uncertainty floor; if the zero-shot model is already over- or underconfident on a dataset, restoring that entropy moves calibration in the wrong direction, as the paper's Caltech101 result shows.

Editorial extensions

If this is right

  • On every entropy-reduced sample, ZAEC provably leaves the complete class ranking, top-1 prediction, and classification accuracy intact, so the calibration gain comes at no accuracy cost.
  • Because ZAEC requires only paired zero-shot and adapted logits and no training or labels, it can be plugged into prompt-, prototype-, cache-, and logit-based TTA methods alike.
  • The diagnostic association implies that calibration damage is concentrated among sharpened samples; the ablation confirms that correcting entropy-increased predictions gives little benefit, supporting the one-sided design.
  • Restoring the full zero-shot entropy floor, rather than a fraction of it, yields the lowest ECE in the evaluated settings, as shown by the monotone improvement in the ablation over restoration strength λ from 0 to 1.
  • Since ZAEC preserves the adapted ranking, it can serve as a drop-in uncertainty layer for downstream selective prediction or human-review triage without changing the decisions being triaged.

Reading between the lines

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

  • If the entropy-reduction signal generalizes beyond CLIP, the same one-sided temperature restoration could apply to any paired frozen-reference/adaptation setup, such as a fine-tuned model compared with its base checkpoint.
  • The Caltech101 boundary case suggests a natural extension the paper leaves open: detect when zero-shot confidence sits systematically below empirical accuracy and make the anchor adaptive so the entropy floor is not restored in that regime.
  • The strong dataset–method correlation between mean relative entropy drop and ECE increase could be repurposed as a label-free monitoring statistic: flag any TTA run with a large mean entropy drop as high-risk for miscalibration before any labels arrive.
  • One-sided minimal temperature scaling can be viewed as projecting the adapted distribution onto the zero-shot entropy level set in forward KL divergence, which may connect ZAEC to risk-control or conformal post-processing theory.
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 manuscript studies calibration degradation caused by test-time adaptation (TTA) of vision-language models. It introduces the notion of prediction-preserving sharpening, in which TTA increases confidence and reduces predictive entropy without changing the top-1 prediction, and shows empirically that larger zero-shot-relative entropy reductions are associated with larger increases in expected calibration error. On the basis of this diagnostic, the authors propose ZAEC, a parameter-free post-hoc method that applies minimal temperature scaling to restore the paired zero-shot entropy, but only for samples whose adapted entropy falls below the zero-shot reference. The method is evaluated on five TTA methods and 15 datasets with two CLIP backbones, and an ablation study tests the effect of correcting different subsets. The paper also documents a boundary case on Caltech101 in which the zero-shot anchor is itself underconfident and ZAEC reverses a beneficial TTA-induced improvement.

Significance. The main contribution is a simple, mechanism-agnostic calibration baseline with no learned parameters and decision-preserving guarantees. The empirical scope is broad (five TTA methods, 15 datasets, two backbones), and the paper is candid about a structural failure case. If the diagnostic association holds beyond the reported settings, zero-shot-relative entropy reduction is a practically useful label-free signal. The paper's own counterexample, however, shows that the anchor can be systematically wrong, and the macro-averaged headline can conceal per-dataset deterioration. The method is likely to be a useful reference baseline, but the central claim needs qualification before the paper can be accepted.

major comments (3)
  1. [Method (Eq. 11) and 'Boundary case on Caltech101'] The central design assumption is that the paired zero-shot entropy is a trustworthy sample-specific uncertainty reference. Section 'Method', Eq. (11), implements this as a one-sided entropy floor: any adapted distribution with H(p^TTA_i) < H(p^ZS_i) is smoothed back up to the zero-shot level. The paper's own boundary case in Figure 4 is a direct counterexample to the generality of this assumption: zero-shot CLIP is underconfident on Caltech101, DPE raises confidence and reduces ECE from 5.07% to 1.69%, and ZAEC restores the zero-shot entropy and increases ECE to 5.38%. This is not a numerical accident of the example; it follows from the one-sided design whenever TTA sharpening corrects zero-shot underconfidence. The abstract and the contribution list state the method without this qualification. The authors should either restrict the central claim to settings where the zero-shot model is not systematically underconfident, add a dataset-level test for anchor reliability, or frame ZAEC explicitly as 'entropy-conservative' rather than 'calibration-improving' in general.
  2. [Experiments, Tables 2 and 3, Figure 4] The headline 'lowest post-hoc macro-average ECE' is an average over datasets and can hide per-dataset degradation that is directly relevant to a user choosing whether to apply ZAEC. For example, on Caltech101 with DPE, raw TTA has ECE 1.69% and ZAEC increases it to 5.38% (Table 3); in Table 2, TDA on ImageNet-R has raw ECE 1.50% and TDA+ZAEC has 4.48%. The paper should report, for each method, the number of datasets on which ZAEC improves or worsens ECE relative to raw TTA, and ideally show the per-dataset reliability diagrams or a scatter plot of raw vs. ZAEC ECE. Without this, the reader cannot tell whether the favorable macro-average reflects consistent small gains or a few large wins offsetting many failures.
  3. [Explaining Calibration Degradation via Zero-Shot-Relative Entropy (Eq. 8, Fig. 2b)] The paper reports r = -0.94 for the dataset-method-level association between mean zero-shot-relative entropy change and ECE change. This single correlation pools samples from multiple TTA methods and datasets; it is not clear how many independent points it is based on, whether the relationship holds separately within each TTA method, or how sensitive it is to the four ImageNet variants versus the fine-grained datasets. Because this correlation is the main evidence for the diagnostic signal, the authors should provide per-method correlations, sample sizes, and confidence intervals (or an equivalent breakdown). If the association is driven by only one or two methods, the motivation for a method-agnostic post-hoc correction is weakened.
minor comments (5)
  1. [Ablation Study] The Ablation Study section opens with a run-on, space-omitted sentence ('All predictions matcheseveryadaptedprediction...'). Rewrite for clarity and separate the definitions of 'All predictions', 'Entropy-increased', and 'Entropy-reduced'.
  2. [Eq. (6) and 'Why logit-range statistics are insufficient'] The quantity Delta^m_R is an average absolute relative deviation; the text 'TPT+SaLS nearly eliminates the range deviation' should say 'on average' to avoid implying per-sample alignment.
  3. [Figure 2(a)] The gray arrows connect SaLS and ZAEC for TPT, but the legend does not indicate the TTA method for the other points; add per-method markers and axis labels for all panels.
  4. [Figure 1 caption] The caption says 'across 11 datasets' while the paper's overall claim is over 15; clarify that Figure 1 covers the second benchmark group only.
  5. [Table 4] The row labels would be clearer as 'Raw TTA (none)', 'All samples', 'Entropy-increased only', and 'Entropy-reduced only (ZAEC)' to align with the ablation text.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: ZAEC is a parameter-free post-hoc rule, and its headline ECE improvements are empirical benchmark outcomes rather than consequences forced by construction. The only mild concern is a same-group benchmark citation and method selection from the authors' own diagnostic, neither of which reduces the derivation to its inputs.

full rationale

The paper's derivation is not circular. ZAEC (Eq. 11) is defined as restoring the paired zero-shot entropy for samples with H(p^TTA) < H(p^ZS); this makes 'entropy restoration' true by definition, but the paper does not present that as a predicted discovery. The load-bearing empirical claims are separate: (1) the diagnostic correlation between zero-shot-relative entropy reduction and ECE increase (Fig. 2b, r = -0.94) is a measured association, not an identity; (2) the headline ECE improvements are benchmark outcomes, not consequences of a fitted parameter. ZAEC has no learned parameters, so the fitted-input-called-prediction pattern does not apply. The method's own boundary case on Caltech101 (DPE raw ECE 5.07% -> 1.69%, DPE+ZAEC 5.38%) shows that restoring zero-shot entropy can hurt calibration, which demonstrates that the good average results are not forced by construction. The only self-citation is the experimental harness: 'Experiments follow the TTABC benchmark (Huang et al. 2026)', developed with overlapping authors; it is a reproducibility framework and is not used to justify the central assumption, so it is at most a minor, non-load-bearing self-citation. The paper also transparently flags the structural limitation: 'future work will explore adaptive anchors for cases where zero-shot underconfidence makes ZAEC overly conservative.' No equation reduces to its input; no fitted parameter is renamed as a prediction.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new free parameters, fitted constants, or physical entities. The method's only reference is the zero-shot entropy, which is computed from the frozen model rather than learned. The core domain assumption is that zero-shot entropy is a trustworthy anchor for calibration, which the authors themselves show can fail in one dataset. The mathematical facts used are standard.

assumptions (3)
  • domain assumption Zero-shot entropy is a valid per-sample uncertainty reference for calibration.
    The entire method rests on using H(p^ZS_i) as the target for entropy restoration. The Caltech101 boundary case shows this assumption can be wrong when the zero-shot model is underconfident.
  • standard math Temperature scaling strictly increases entropy and preserves class rankings for any non-constant logit vector.
    Proposition 1 in Section 'Method'; this is a standard property of softmax and convexity.
  • domain assumption Top-label ECE is an appropriate measure of calibration for the downstream decisions considered.
    The paper uses top-label ECE exclusively as the calibration metric; other definitions could rank methods differently.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Respect Your Zero-Shot Uncertainty: Conservative Calibration for Test-Time-Adapted Vision-Language Models." pith.science (2026). https://pith.science/paper/NOK35E26

@misc{pith2026260805945,
  author       = {Pith},
  title        = {Pith review of: Respect Your Zero-Shot Uncertainty: Conservative Calibration for Test-Time-Adapted Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NOK35E26}},
  note         = {Machine review of arXiv:2608.05945}
}
read the original abstract

Test-time adaptation (TTA) can improve the recognition accuracy of vision-language models under distribution shift, but often degrades calibration, making predictive confidence unreliable for downstream decision-making. Many existing label-free calibration approaches are either coupled to prompt optimization or rely on logit-range statistics that provide only a coarse characterization of the predictive distribution. We show that TTA can increase confidence and reduce entropy even when the top-1 prediction and its correctness remain unchanged, a failure mode we term prediction-preserving sharpening. Across diverse TTA methods and benchmarks, larger entropy reductions relative to paired zero-shot predictions are associated with greater increases in Expected Calibration Error (ECE). On entropy-reduced samples, confidence gains also tend to exceed accuracy gains. Based on these findings, we propose Zero-Shot-Anchored Entropy Calibration (ZAEC), a label-free post-hoc method that uses zero-shot entropy as a sample-specific uncertainty reference. ZAEC selectively restores the zero-shot entropy of sharpened predictions through minimal temperature scaling while leaving all other predictions unchanged. It requires no labeled calibration data or learned parameters and preserves class rankings and classification accuracy. Across five TTA methods and 15 datasets, ZAEC achieves the lowest post-hoc macro-average ECE on ViT-B/16, with consistent gains on RN50.

Figures

Figures reproduced from arXiv: 2608.05945 by the authors.

Figure 1
Figure 1. Expected Calibration Error (ECE) of raw TTA, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Diagnostic analyses of TTA-induced calibration degradation. (a) Exact zero-shot logit-range matching can retain high calibration error. Gray arrows connect SaLS and ZAEC applied to the same TPT predictions. (b) At the dataset–method level, stronger zero-shot-relative entropy reduction is associated with larger ECE degradation. (c) Among entropy-reduced samples, stronger entropy reduction corresponds to a larger conf… view at source ↗
Figure 3
Figure 3. Average ECE and predictive entropy on four ImageNet natural distribution-shift datasets using CLIP RN50. ZAEC restores mean predictive entropy toward the zero-shot reference and achieves the lowest average ECE for each evaluated TTA method. 0.00.10.20.30.40.50.60.70.80.91.0 Confidence 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Accuracy (a) Zero-shot Caltech101 ECE=5.07% Outputs Gap 0.00.10.20.30.40.50.60.70.80.91.0… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Boundary case of ZAEC on Caltech101. Implementation details. We evaluate CLIP with ViT￾B/16 (Dosovitskiy et al. 2021) and RN50 (He et al. 2016), reporting ViT-B/16 results in the main paper and com￾plete RN50 results in the appendix. Experiments follow the TTABC benchm…
Figure 5
Figure 5. Figure 5: Reliability diagrams on ImageNet-A using CLIP ViT-B/16. Rows correspond to raw TTA, TTA+SaLS, and TTA+ZAEC, while columns correspond to TPT, TDA, and DPE. Predictions are grouped into 10 confidence bins. Blue bars show empirical bin accuracy, and hatched regions indi￾c…
Figure 6
Figure 6. Figure 6: Ablation of entropy-restoration strength on DTD and ImageNet-V2 with CLIP ViT-B/16. H(p ZS i ), we define the target entropy as H target i (λ) = H [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

91 extracted references · 40 canonical work pages

  1. [1]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  2. [2]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  3. [3]

    ICLR , year=

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , author=. ICLR , year=

  4. [4]

    arXiv preprint arXiv:2303.15361 , year=

    A comprehensive survey on test-time adaptation under distribution shifts , author=. arXiv preprint arXiv:2303.15361 , year=

    arXiv

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    The illusion of progress? a critical look at test-time adaptation for vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  6. [6]

    arXiv preprint arXiv:2006.10726 , year=

    Tent: Fully test-time adaptation by entropy minimization , author=. arXiv preprint arXiv:2006.10726 , year=

    arXiv 2006

  7. [7]

    Entropy is not Enough for Test-Time Adaptation: From the Perspective of Disentangled Factors , author=

  8. [8]

    Advances in neural information processing systems , volume=

    Memo: Test time robustness via adaptation and augmentation , author=. Advances in neural information processing systems , volume=

Show all 91 references
  1. [9]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Continual test-time domain adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  2. [10]

    CoRR , volume =

    Schneider, Steffen and Rusak, Evgenia and Eck, Luisa and Bringmann, Oliver and Brendel, Wieland and Bethge, Matthias , title =. CoRR , volume =

  3. [11]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Gong, Taesik and Jeong, Jongheon and Kim, Taewon and Kim, Yewon and Shin, Jinwoo and Lee, Sung-Ju , title =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  4. [12]

    arXiv preprint arXiv:2302.12400 , year=

    Towards stable test-time adaptation in dynamic wild world , author=. arXiv preprint arXiv:2302.12400 , year=

    arXiv

  5. [13]

    International journal of computer vision , volume=

    Learning to prompt for vision-language models , author=. International journal of computer vision , volume=. 2022 , publisher=

    2022

  6. [14]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Conditional prompt learning for vision-language models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  7. [15]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Maple: Multi-modal prompt learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  8. [16]

    Advances in Neural Information Processing Systems , volume=

    Test-time prompt tuning for zero-shot generalization in vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  9. [17]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Diverse data augmentation with diffusions for effective test-time prompt tuning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  10. [18]

    Advances in Neural Information Processing Systems , volume=

    Historical test-time prompt tuning for vision foundation models , author=. Advances in Neural Information Processing Systems , volume=

  11. [19]

    Hasegawa-Johnson and Yingzhen Li and Chang D

    Hee Suk Yoon and Eunseop Yoon and Joshua Tian Jin Tee and Mark A. Hasegawa-Johnson and Yingzhen Li and Chang D. Yoo , booktitle=. C-

  12. [20]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    O-tpt: Orthogonality constraints for calibrating test-time prompt tuning in vision-language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  13. [21]

    Perera Miriya Thanthrige and Ranga Rodrigo and Muhammad Haris Khan , booktitle=

    Shihab Aaqil Ahamed and Udaya Sampath K. Perera Miriya Thanthrige and Ranga Rodrigo and Muhammad Haris Khan , booktitle=. A-

  14. [22]

    Advances in Neural Information Processing Systems , volume=

    Align your prompts: Test-time prompting with distribution alignment for zero-shot generalization , author=. Advances in Neural Information Processing Systems , volume=

  15. [23]

    2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages=

    Just shift it: Test-time prototype shifting for zero-shot generalization with vision-language models , author=. 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , pages=. 2025 , organization=

    2025

  16. [24]

    Advances in Neural Information Processing Systems , volume=

    Dual prototype evolving for test-time generalization of vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  17. [25]

    arXiv preprint arXiv:2412.02837 , year=

    BATCLIP: Bimodal Online Test-Time Adaptation for CLIP , author=. arXiv preprint arXiv:2412.02837 , year=

    arXiv

  18. [26]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Panda: Test-Time Adaptation with Negative Data Augmentation , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  19. [27]

    arXiv preprint arXiv:2305.18010 , year=

    Test-time adaptation with clip reward for zero-shot generalization in vision-language models , author=. arXiv preprint arXiv:2305.18010 , year=

    arXiv

  20. [28]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Efficient test-time adaptation of vision-language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  21. [29]

    Advances in Neural Information Processing Systems , volume=

    Boostadapter: Improving vision-language test-time adaptation via regional bootstrapping , author=. Advances in Neural Information Processing Systems , volume=

  22. [30]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , year=

    Dual Memory Networks: A Versatile Adaptation Approach for Vision-Language Models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , year=

  23. [31]

    arXiv preprint arXiv:2409.19375 , year=

    Dota: Distributional test-time adaptation of vision-language models , author=. arXiv preprint arXiv:2409.19375 , year=

  24. [32]

    , author=

    Online Gaussian Test-Time Adaptation of Vision-Language Models. , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  25. [33]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Bayesian test-time adaptation for vision-language models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  26. [34]

    Advances in Neural Information Processing Systems , volume=

    Frustratingly easy test-time adaptation of vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  27. [35]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Calip: Zero-shot enhancement of clip with parameter-free attention , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  28. [36]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    On the test-time zero-shot generalization of vision-language models: Do we really need prompt learning? , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  29. [37]

    European Conference on Computer Vision , pages=

    Online zero-shot classification with clip , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  30. [38]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    R-tpt: Improving adversarial robustness of vision-language models through test-time prompt tuning , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  31. [39]

    arXiv preprint arXiv:2412.18303 , year=

    Efficient and context-aware label propagation for zero-/few-shot training-free adaptation of vision-language model , author=. arXiv preprint arXiv:2412.18303 , year=

    arXiv

  32. [40]

    arXiv preprint arXiv:2501.16404 , year=

    Dynaprompt: Dynamic test-time prompt tuning , author=. arXiv preprint arXiv:2501.16404 , year=

    arXiv

  33. [41]

    arXiv preprint arXiv:2411.03687 , year=

    Beyond model adaptation at test time: A survey , author=. arXiv preprint arXiv:2411.03687 , year=

    arXiv

  34. [42]

    arXiv preprint arXiv:1912.02781 , year=

    Augmix: A simple data processing method to improve robustness and uncertainty , author=. arXiv preprint arXiv:1912.02781 , year=

    1912 arXiv

  35. [43]

    International conference on machine learning , pages=

    Do imagenet classifiers generalize to imagenet? , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  36. [44]

    Advances in Neural Information Processing Systems , pages=

    Learning Robust Global Representations by Penalizing Local Predictive Power , author=. Advances in Neural Information Processing Systems , pages=

  37. [45]

    CVPR , year=

    Natural Adversarial Examples , author=. CVPR , year=

  38. [46]

    ICCV , year=

    The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization , author=. ICCV , year=

  39. [47]

    3D Object Representations for Fine-Grained Categorization , year=

    Krause, Jonathan and Stark, Michael and Deng, Jia and Fei-Fei, Li , booktitle=. 3D Object Representations for Fine-Grained Categorization , year=

  40. [48]

    European Conference on Computer Vision , year =

    Food-101 -- Mining Discriminative Components with Random Forests , author =. European Conference on Computer Vision , year =

  41. [49]

    Fine-Grained Visual Classification of Aircraft , author=

  42. [50]

    2012 IEEE conference on computer vision and pattern recognition , year=

    Cats and dogs , author=. 2012 IEEE conference on computer vision and pattern recognition , year=

    2012

  43. [51]

    2008 Sixth Indian conference on computer vision, graphics & image processing , pages=

    Automated flower classification over a large number of classes , author=. 2008 Sixth Indian conference on computer vision, graphics & image processing , pages=. 2008 , organization=

    2008

  44. [52]

    2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition , title=

    J. 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition , title=. 2010 , month=

    2010

  45. [53]

    Proceedings of the IEEE conference on computer vision and pattern recognition , year=

    Describing textures in the wild , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , year=

  46. [54]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , year=

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification , author=. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , year=

  47. [55]

    arXiv preprint arXiv:1212.0402 , year=

    Ucf101: A dataset of 101 human actions classes from videos in the wild , author=. arXiv preprint arXiv:1212.0402 , year=

    arXiv

  48. [56]

    arXiv preprint arXiv:1903.12261 , year=

    Benchmarking neural network robustness to common corruptions and perturbations , author=. arXiv preprint arXiv:1903.12261 , year=

    1903 arXiv

  49. [57]

    arXiv preprint arXiv:1807.03748 , year=

    Representation learning with contrastive predictive coding , author=. arXiv preprint arXiv:1807.03748 , year=

    arXiv

  50. [58]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Robust test-time adaptation in dynamic scenarios , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  51. [59]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Realistic test-time adaptation of vision-language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  52. [60]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Open-vocabulary semantic segmentation with mask-adapted clip , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  53. [61]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Cora: Adapting clip for open-vocabulary detection with region prompting and anchor pre-matching , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  54. [62]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Vision-language models for vision tasks: A survey , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2024 , publisher=

    2024

  55. [63]

    International conference on machine learning , pages=

    Scaling up visual and vision-language representation learning with noisy text supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  56. [64]

    International conference on machine learning , pages=

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  57. [65]

    2009 IEEE conference on computer vision and pattern recognition , pages=

    Imagenet: A large-scale hierarchical image database , author=. 2009 IEEE conference on computer vision and pattern recognition , pages=. 2009 , organization=

    2009

  58. [66]

    European Conference on Computer Vision , pages=

    Robust calibration of large vision-language adapters , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  59. [67]

    International conference on machine learning , pages=

    On calibration of modern neural networks , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  60. [68]

    Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    A survey on multimodal large language models for autonomous driving , author=. Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

  61. [69]

    Transportation Research Part C: Emerging Technologies , volume=

    V2x-vlm: End-to-end v2x cooperative autonomous driving through large vision-language models , author=. Transportation Research Part C: Emerging Technologies , volume=. 2026 , publisher=

    2026

  62. [70]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Clip-driven universal model for organ segmentation and tumor detection , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  63. [71]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Towards Calibrating Prompt Tuning of Vision-Language Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  64. [72]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    SoC: Semantic Orthogonal Calibration for Test-Time Prompt Tuning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  65. [73]

    arXiv preprint arXiv:2402.04655 , year=

    Open-vocabulary calibration for fine-tuned clip , author=. arXiv preprint arXiv:2402.04655 , year=

    arXiv

  66. [74]

    Advances in large margin classifiers , volume=

    Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods , author=. Advances in large margin classifiers , volume=. 1999 , publisher=

    1999

  67. [75]

    Icml , volume=

    Obtaining calibrated probability estimates from decision trees and naive bayesian classifiers , author=. Icml , volume=

  68. [76]

    2005 , publisher=

    Algorithmic learning in a random world , author=. 2005 , publisher=

    2005

  69. [77]

    Journal of the American Statistical Association , volume=

    Distribution-free predictive inference for regression , author=. Journal of the American Statistical Association , volume=. 2018 , publisher=

    2018

  70. [78]

    International Conference on Machine Learning , pages=

    Trainable calibration measures for neural networks from kernel mean embeddings , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    2018

  71. [79]

    Advances in Neural Information Processing Systems , volume=

    Soft calibration objectives for neural networks , author=. Advances in Neural Information Processing Systems , volume=

  72. [80]

    Advances in neural information processing systems , volume=

    On mixup training: Improved calibration and predictive uncertainty for deep neural networks , author=. Advances in neural information processing systems , volume=

  73. [81]

    arXiv preprint arXiv:2404.16168 , year=

    The Over-Certainty Phenomenon in Modern Test-Time Adaptation Algorithms , author=. arXiv preprint arXiv:2404.16168 , year=

    arXiv

  74. [82]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    The devil is in the margin: Margin-based label smoothing for network calibration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  75. [83]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Test-time distillation for continual model adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  76. [84]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Neural Collapse in Test-Time Adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  77. [85]

    27th European Conference on Artificial Intelligence , volume=

    Cliff: Leveraging Ambiguous Samples for Enhanced Test-Time Adaptation , author =. 27th European Conference on Artificial Intelligence , volume=. 2024 , doi =

    2024

  78. [86]

    Proceedings of the AAAI Conference on Artificial Intelligence , pages=

    Moetta: Test-time adaptation under mixed distribution shifts with moe-layernorm , author=. Proceedings of the AAAI Conference on Artificial Intelligence , pages=

  79. [87]

    arXiv preprint arXiv:2606.14299 , year=

    What Drives Test-Time Adaptation for CLIP? A Controlled Empirical Study from an Update Perspective , author=. arXiv preprint arXiv:2606.14299 , year=

    arXiv

  80. [88]

    International Conference on Intelligent Computing , pages=

    Test-Time Adaptation via Dynamic Historical Knowledge Vector Fusion , author=. International Conference on Intelligent Computing , pages=. 2025 , organization=

    2025

  81. [89]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Dynamic Logits Adjustment and Exploration for Test-Time Adaptation in Vision Language Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  82. [90]

    Proceedings of the AAAI conference on artificial intelligence , year=

    Obtaining well calibrated probabilities using bayesian binning , author=. Proceedings of the AAAI conference on artificial intelligence , year=

  83. [91]

    2004 conference on computer vision and pattern recognition workshop , pages=

    Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories , author=. 2004 conference on computer vision and pattern recognition workshop , pages=. 2004 , organization=

    2004

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.