Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Mahalanobis++: Improving OOD Detection via Feature Normalization

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read L2-normalizing pre-logit features before the Mahalanobis score makes OOD detection consistently stronger across 44 ImageNet models.

desk verdict Simple post-hoc fix that works on 41 of 44 models, but 'consistent' and 'all considered cases' are overstatements that need aligning with the paper's own tables. read the letter →

arxiv 2505.18032 v1 pith:MJL3D2NZ submitted 2025-05-23 cs.LG cs.CV

classification cs.LGcs.CV
keywords out-of-distributiondetectionMahalanobisdistancefeaturenormalizationL2pre-logitfeaturesGaussianassumptionImageNetevaluationpost-hocmethods
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

Post-hoc out-of-distribution (OOD) detection has to work on already-trained classifiers. This paper identifies why the Mahalanobis distance, one of the most effective post-hoc detectors, is inconsistent across models: the pre-logit features it relies on violate the assumed Gaussian model, with feature norms that vary far more across and within classes than a shared-covariance Gaussian would produce, and small-norm OOD samples being systematically scored as in-distribution. The proposed fix, Mahalanobis++, normalizes each pre-logit feature vector to unit length before estimating class means, the shared covariance matrix, and the final score. Across 44 ImageNet models the normalization improves average false positive rate by 7.6 points over the unnormalized Mahalanobis baseline, outperforms the previous best post-hoc method by about 7 points on average, and removes the fragility on simple noise distributions for several models. If the claim holds, practitioners gain a hyperparameter-free modification that works across architectures and pretraining schemes without retraining.

What carries the argument

The machinery is the pre-logit feature vector $\phi(x)$, the Mahalanobis distance $d_{\mathrm{Maha}}(x,\hat{\mu}_c)=(\phi(x)-\hat{\mu}_c)^T\hat{\Sigma}^{-1}(\phi(x)-\hat{\mu}_c)$ to the class mean $\hat{\mu}_c$, and the replacement of $\phi(x)$ by $\hat{\phi}(x)=\phi(x)/\|\phi(x)\|_2$ in every step: class-mean estimation, shared-covariance estimation, and scoring. This projects all features onto the unit sphere and discards scale information. The paper motivates the replacement with a concentration lemma showing that truly Gaussian high-dimensional features have concentrated norms, with QQ plots and a variance-alignment score showing that real features are heavy-tailed and class-dependent, and with a documented correlation between feature norm and the unnormalized Mahalanobis score that normalization breaks.

What would settle it

On a model whose feature norms still correlate with the Mahalanobis++ score after normalization, compute the rank correlation between original feature norm and the normalized OOD score over in- and out-of-distribution test sets; the paper's mechanism predicts near-zero correlation, so a large residual correlation would falsify the claim that normalization decouples the score from feature scale.

Watch

Extended reading notes

Core claim

The paper's central claim is that L2-normalizing pre-logit features before computing the Mahalanobis distance makes a brittle OOD detector consistently strong, because the normalization makes the empirical feature distribution better satisfy the class-conditional Gaussian model with shared covariance that the Mahalanobis score implicitly assumes. Evidence is the 44-model ImageNet study and CIFAR experiments: Mahalanobis++ improves FPR over unnormalized Mahalanobis in 41 of 44 models, with average gains of 7.6 FPR points on the five-dataset OpenOOD suite and 10.9 points on the cleaned NINCO dataset. The only models that do not benefit are those whose feature norms are already tightly concentrated—notably two augreg-trained ViTs—which is consistent with the explanation rather than a refutation of it.

Load-bearing premise

Normalizing away each sample's feature length removes only harmless scale information while preserving the directional pattern that separates classes; if a model carries its OOD signal in the feature norm itself, this fix can backfire.

Editorial extensions

If this is right

  • Mahalanobis++ outperforms the conventional Mahalanobis distance in 41 of 44 ImageNet models and beats the strongest previous post-hoc baseline, ViM, by about 7 average FPR points.
  • Because it is post-hoc and hyperparameter-free, the method applies directly to any frozen pretrained classifier, including models whose pretraining data or procedure is not publicly available.
  • Normalization repairs the known brittleness of Mahalanobis-based detectors on far-OOD noise distributions: for several representative models the number of failed unit tests drops from 10-14 to 0.
  • The gain is largest on NINCO, where contamination with in-distribution images is removed, with an average improvement of 10.9 FPR points and a new best FPR of 18.4%.
  • On CIFAR100 the recipe improves Mahalanobis in every tested model, but by smaller margins, consistent with weaker norm pathology at smaller scale.

Reading between the lines

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

  • The pre-training feature-norm diagnostics could serve as an OOD-data-free early check: a model whose class norms are spread or correlated with the Mahalanobis score is a candidate for normalization, while a model like the augreg ViT is not.
  • Any distance- or norm-based OOD score may inherit the same scale confound, so L2 normalization is likely to help other post-hoc detectors, not only the Mahalanobis family; the paper compares against but does not make this claim.
  • Because well-behaved features in this paper come from a specific augmentation-and-regularization recipe whose mechanism is left unexplained, the results hint that training schemes could deliberately induce norm-regulated features and make post-hoc OOD detection uniformly reliable.
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 / 6 minor

Summary. The paper proposes Mahalanobis++, a post-hoc OOD detection method that applies l2-normalization to pre-logit features before estimating and using the Mahalanobis distance. The authors argue that feature norm variations violate the Gaussian shared-covariance assumption of the Mahalanobis score, and that projecting features onto the unit sphere reduces this violation and decouples the score from feature norm. They evaluate the method on 44 ImageNet checkpoints and several CIFAR models, reporting an average FPR improvement of 7.6 points over standard Mahalanobis on OpenOOD datasets, 10.9 points on the cleaned NINCO subset, and state-of-the-art average performance relative to a broad set of baselines. The paper also claims that normalization remedies the brittleness of Mahalanobis-based detectors on noise unit tests.

Significance. If the empirical claims are properly qualified, this is a practically valuable contribution: the method is simple, adds no fitted parameters, and is applicable to any pretrained model. The evidence base is unusually strong for a post-hoc OOD paper—44 diverse checkpoints, five OOD datasets, multiple architectures, and three complementary diagnostic analyses (norm distributions, QQ plots, variance-alignment scores). The hypothesis was formed on one model and validated on held-out checkpoints, and code is provided, which increases confidence and reproducibility. The main weakness is that the abstract and several central statements overgeneralize: the paper's own results show three models where normalization degrades FPR, and the full unit-test table shows many models still fail a substantial number of noise tests. These overclaims, if corrected, do not undermine the core averaged improvement but they do require a revised, conditional formulation of the method's benefits.

major comments (3)
  1. [Abstract, §4, Table 4] The claim that l2-normalization 'improves the conventional Mahalanobis distance-based approaches significantly and consistently' is contradicted by the paper's own Table 4: three models (ViT-B16-In21k-augreg, ViT-L16-In21k-augreg, ResNet50) show worse FPR after normalization (25.7→28.3, 21.0→23.9, 49.5→52.0). Appendix D attributes these exceptions to already well-behaved feature norms, which implies the benefit is conditional on the degree of norm variance. The abstract, Section 4, and conclusion should state this boundary explicitly rather than claiming consistency.
  2. [§5, Table 5, Appendix Table 17] The main text states that normalization 'remedies the brittleness' of Mahalanobis-based detectors and Table 5 highlights three models with perfect 0/17 unit-test outcomes, but the full results in Table 17 show that many models still fail a substantial number of unit tests after normalization (e.g., ConvNeXt-B: 15, DeiT3-B16: 15, DeiT3-L16: 8, DeiT3-S16-In21k: 11, Mixer-B16-In21k: 10), and for ResNet101 and ResNet50 the failure count increases from 0 to 1. The robustness claim should be rephrased as 'reduces the number of failures for most models' and the full distribution of failure counts should be summarized in the main text.
  3. [§3.3, Figure 10, §4] The paper treats feature norm as a pure confounder, but Figure 10 shows that for several models (e.g., ViT-B16-orig, ViT-B16-miil, ViT-B16-clip) natural OOD samples from NINCO have larger feature norms than ID samples, meaning the norm carries class-discriminative signal that l2-normalization discards. The diagnostic evidence for the confounder claim is built primarily on SwinV2 and models with small OOD norms; it does not establish that discarding the norm is always safe. The paper should explicitly state that normalization is beneficial when the norm-OOD-score correlation is confounded (as in Figure 5) and should report conditions under which it helps, e.g., using the variance-alignment metric of Table 7 or the norm overlap of Figure 10 as a predictor.
minor comments (6)
  1. [Abstract] There is a typo: 'safety-critial' should be 'safety-critical'.
  2. [§3.1] The reference 'Lee et al. (2018a)' for the Gaussian assumption should likely be 'Lee et al. (2018b)', since the cited 2018a paper is about confidence calibration rather than the Mahalanobis framework.
  3. [§4] The subsection 'Improved normality' claims that normalization makes features 'more aligned with a normal distribution,' but after l2-normalization the features lie on the unit sphere, which cannot be multivariate Gaussian. The QQ-plot evidence on centered, standardized coordinates is an indirect diagnostic; the text should not literally assert that normalized features are better modeled as a shared-covariance Gaussian.
  4. [Table 5] The criteria for selecting the three reported models (ConvNeXtV2, SwinV2, ViT-CLIP) should be stated; without this, the selection appears arbitrary and potentially cherry-picked relative to the full results in Table 17.
  5. [§5, ImageNet] The sentence 'In 30/44 cases, the best performing method is Mahalanobis++ (in 6/44 cases it is relative Mahalanobis++)' should clarify what happens in the remaining 8 cases, e.g., whether they are ties or another method.
  6. [Figure 6] The vertical line labeled 'True sample norm' is unclear; the caption should describe that this marks the unmodified feature norm of the OOD samples.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the normalization transform is fixed and parameter-free, and the claimed improvements are measured against external OOD benchmarks.

full rationale

The central claim is an empirical comparison, not a derivation from its own inputs. The method (Eq. 6) is a fixed, hyperparameter-free L2-normalization applied to pre-logit features before computing class means and a shared covariance matrix (Eqs. 1-2). No parameter is fitted to OOD detection outcomes, and the threshold is set by fixing TPR=95% on in-distribution data, so the reported FPR values are not forced by construction. Lemma 3.1 and Eq. 5 are derived from first principles in the appendix and serve only as motivation; they do not encode the FPR improvements. The 44 checkpoints are publicly available pretrained models, and the evaluation uses the external OpenOOD and NINCO benchmarks. The paper self-cites Bitterwolf et al. (2023) for the NINCO benchmark and unit tests, and Mueller & Hein (2024) for related training-hyperparameter analysis, but these citations are data/benchmark sources and pointers to prior work, not load-bearing premises that reduce the central result to a self-citation. The abstract's wording 'consistently' is somewhat overstated given that three models degrade (Table 4), but this is a correctness/scope caveat rather than circularity. Overall, the claimed improvement over unnormalized Mahalanobis distance is a measured, externally grounded result.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

Mahalanobis++ is a hyperparameter-free method; the ledger's free_parameters list is empty. The key assumptions are the Gaussian model inherited from Lee et al. (2018), the empirical claim that normalization preserves discriminative direction while discarding norm, and standard probabilistic tools used in the proofs. No new entities are postulated.

assumptions (4)
  • domain assumption Pre-logit features follow a class-conditional Gaussian distribution with a shared covariance matrix (Lee et al., 2018).
    The method is built on this generative model; the paper shows it is violated for some models and argues normalization restores it. Invoked in Section 3.1 and Section 4.
  • ad hoc to paper L2-normalization preserves the directional information needed for OOD discrimination and discards only the confounding norm.
    The core mechanism of the fix. Supported empirically, but no theoretical guarantee, and contradicted for three models in Table 4. Invoked in Section 4, Eq. (6).
  • domain assumption Sample covariance estimated on the training set is a reliable proxy for the true shared covariance in high dimension.
    The Mahalanobis distance requires inverting the covariance (Eq. 3); no regularization or conditioning analysis is reported.
  • standard math Chebyshev's inequality and standard high-dimensional concentration and spherical moment computations.
    Used in Lemma 3.1 and Appendix C; proofs are provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mahalanobis++: Improving OOD Detection via Feature Normalization." pith.science (2026). https://pith.science/paper/MJL3D2NZ

@misc{pith2026250518032,
  author       = {Pith},
  title        = {Pith review of: Mahalanobis++: Improving OOD Detection via Feature Normalization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MJL3D2NZ}},
  note         = {Machine review of arXiv:2505.18032}
}
abstract

Detecting out-of-distribution (OOD) examples is an important task for deploying reliable machine learning models in safety-critial applications. While post-hoc methods based on the Mahalanobis distance applied to pre-logit features are among the most effective for ImageNet-scale OOD detection, their performance varies significantly across models. We connect this inconsistency to strong variations in feature norms, indicating severe violations of the Gaussian assumption underlying the Mahalanobis distance estimation. We show that simple $\ell_2$-normalization of the features mitigates this problem effectively, aligning better with the premise of normally distributed data with shared covariance matrix. Extensive experiments on 44 models across diverse architectures and pretraining schemes show that $\ell_2$-normalization improves the conventional Mahalanobis distance-based approaches significantly and consistently, and outperforms other recently proposed OOD detection methods.

Figures

Figures reproduced from arXiv: 2505.18032 by the authors.

Figure 1
Figure 1. Normalizing features improves OOD detection with the Mahalanobis distance consistently. Shown is the difference in false-positive rate at true positive rate of 95% between unnormal￾ized and normalized features for 44 ImageNet models, averaged over five OOD datasets of the OpenOOD benchmark. genuine ID samples to pass through normally. OOD de￾tection methods are commonly divided into methods that require modification… view at source ↗
Figure 2
Figure 2. Mahalanobis++: We illustrate how to improve Mahalanobis-based OOD detection. Left: For unnormalized features, assuming a shared covariance matrix for all classes leads to suboptimal OOD detection (bottom) with the Mahalanobis score. Center: Normalizing the features, i.e. projecting them onto the unit sphere mitigates this problem effectively. Right: After normalization, the fit of the shared covariance matrix is tig… view at source ↗
Figure 3
Figure 3. The feature norms vary strongly across and within classes. Left: We simulate how the feature norms per class would be distributed if they were sampled from Gaussians with the means and covariance matrix used for the Mahalanobis distance estima￾tion. Right: The actual feature norm distribution observed in practice. Both the average norms across classes and the norms within each class vary much stronger than expected.… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: QQ-plot: ℓ2−normalization helps transform the fea￾tures to be more aligned with a normal distribution. For a SwinV2 and DeiT3 model (where the feature norms vary strongly across and within classes) normalization shifts the distribution towards a Gaussian (black line). …
Figure 5
Figure 5. Figure 5: The feature norm correlates with the Mahalanobis score for SwinV2-B: Left: The smaller the feature norm, the smaller the Mahalanobis OOD score sMaha, irrespective of whether a sample is ID or not. OOD samples with small feature norms are systematically classified as ID…
Figure 6
Figure 6. Figure 6 [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Mahalanobis++ resolves feature-norm dependency of Mahalanobis score. With unnormalized features, OOD samples with small pre-logit feature norm were systematically identified as ID, but after normalization, OOD samples with small feature norm are rightfully detected as …
Figure 8
Figure 8. Figure 8: For most models, the feature norms vary strongly across and within classes. The same plot as the ”observed” part of [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: QQ-plot: ℓ2−normalization helps transform the features to be more aligned with a normal distribution. Normalized features in green, unnormalized features in blue. For a SwinV2, DeiT3 and ViT-augreg2, the feature norms vary strongly across classes (see e.g [PITH_FULL_I…
Figure 10
Figure 10. Figure 10: Feature norm distribution. In contrast to previous work (e.g., (Park et al., 2023b)), we find that the feature norm of natural OOD samples (NINCO in green) is often larger than that of ID samples (orange). Far-OOD data, like noise distributions, tend to have lower fea…
Figure 11
Figure 11. Figure 11: Mahalanobis++ aligns class-variances. We report the distribution of the variances of the train features for each class along three directions: 1) a random direction, 2) a large eigendirection, 3) a small eigendirection. For each class, we compute the variance divided …
Figure 12
Figure 12. Figure 12: We plot the FPR with Mahalanobis++ against the FPR with the conventional Mahalanobis score averaged over the five OpenOOD datasets. With three minor exceptions, Mahalanobis++ improves OOD detection performance for all models. In particular, it significantly improves a…

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. Fine-Grained Open-Set Fault Diagnosis via Metric-Guided Time-Frequency Configuration Selection and Class-Specific Autoencoders

    eess.SP 2026-07 conditional novelty 5.0 of 10

    A pipeline that screens STFT settings by Silhouette score, trains one autoencoder per severity class, and rejects via latent-box plus reconstruction-error checks reports H-scores of 0.9924 (CWRU) and 0.9509 (PU) for f...

Reference graph

Works this paper leans on

56 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    B., Belkhir, N., Popescu, S., Manzanera, A., and Franchi, G

    Ammar, M. B., Belkhir, N., Popescu, S., Manzanera, A., and Franchi, G. NECO : NE ural collapse based out-of-distribution detection. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=9ROuKblmi7

  3. [3]

    and Kamnitsas, K

    Anthony, H. and Kamnitsas, K. On the use of mahalanobis distance for out-of-distribution detection with neural networks for medical imaging. In Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, pp.\ 136--146. Springer Nature Switzerland, 2023. doi:10.1007/978-3-031-44336-7_14. URL https://doi.org/10.1007

  4. [4]

    In or out? fixing imagenet out-of-distribution detection evaluation

    Bitterwolf, J., Mueller, M., and Hein, M. In or out? fixing imagenet out-of-distribution detection evaluation. In ICML, 2023. URL https://proceedings.mlr.press/v202/bitterwolf23a.html

  5. [5]

    Describing textures in the wild

    Cimpoi, M., Maji, S., Kokkinos, I., Mohamed, S., and Vedaldi, A. Describing textures in the wild. In CVPR, 2014

  6. [6]

    Extremely simple activation shaping for out-of-distribution detection

    Djurisic, A., Bozanic, N., Ashok, A., and Liu, R. Extremely simple activation shaping for out-of-distribution detection. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=ndYXTEL6cZz

  7. [7]

    A framework for benchmarking class-out-of-distribution detection and its application to imagenet

    Galil, I., Dabbah, M., and El-Yaniv, R. A framework for benchmarking class-out-of-distribution detection and its application to imagenet. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=Iuubb9W6Jtk

  8. [8]

    Gia, T. L. and Ahn, J. Understanding normalization in contrastive representation learning and out-of-distribution detection. ArXiv, abs/2312.15288, 2023. URL https://api.semanticscholar.org/CorpusID:266550859

Show all 56 references
  1. [9]

    Haas, J., Yolland, W., and Rabus, B. T. Linking neural collapse and l2 normalization with improved out-of-distribution detection in deep neural networks. Transactions on Machine Learning Research, 2023. ISSN 2835-8856. URL https://openreview.net/forum?id=fjkN5Ur2d6

  2. [10]

    Haas, J., Yolland, W., and Rabus, B. T. Exploring simple, high quality out-of-distribution detection with l2 normalization. Transactions on Machine Learning Research, 2024. ISSN 2835-8856. URL https://openreview.net/forum?id=daX2UkLMS0

  3. [11]

    Why ReLU networks yield high-confidence predictions far away from the training data and how to mitigate the problem

    Hein, M., Andriushchenko, M., and Bitterwolf, J. Why ReLU networks yield high-confidence predictions far away from the training data and how to mitigate the problem. In CVPR, 2019

  4. [12]

    and Gimpel, K

    Hendrycks, D. and Gimpel, K. A baseline for detecting misclassified and out-of-distribution examples in neural networks. In ICLR, 2017. URL https://openreview.net/forum?id=Hkg4TI9xl

  5. [13]

    Scaling out-of-distribution detection for real-world settings

    Hendrycks, D., Basart, S., Mazeika, M., Zou, A., Kwon, J., Mostajabi, M., Steinhardt, J., and Song, D. Scaling out-of-distribution detection for real-world settings. In ICML, 2022

  6. [14]

    Oodformer: Out-of-distribution detection transformer, 07 2021

    Koner, R., Sinhamahapatra, P., Roscher, K., Günnemann, S., and Tresp, V. Oodformer: Out-of-distribution detection transformer, 07 2021

  7. [15]

    Openimages: A public dataset for large-scale multi-label and multi-class image classification

    Krasin, I., Duerig, T., Alldrin, N., Ferrari, V., Abu-El-Haija, S., Kuznetsova, A., Rom, H., Uijlings, J., Popov, S., Kamali, S., Malloci, M., Pont-Tuset, J., Veit, A., Belongie, S., Gomes, V., Gupta, A., Sun, C., Chechik, G., Cai, D., Feng, Z., Narayanan, D., and Murphy, K. O...

  8. [16]

    Learning multiple layers of features from tiny images

    Krizhevsky, A. Learning multiple layers of features from tiny images. Technical report, 2009

  9. [17]

    and Yang, X

    Le, Y. and Yang, X. S. Tiny imagenet visual recognition challenge. 2015. URL https://api.semanticscholar.org/CorpusID:16664790

  10. [18]

    Gradient-based learning applied to document recognition

    LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86 0 (11): 0 2278--2324, 1998. doi:10.1109/5.726791

  11. [19]

    Training confidence-calibrated classifiers for detecting out-of-distribution samples

    Lee, K., Lee, H., Lee, K., and Shin, J. Training confidence-calibrated classifiers for detecting out-of-distribution samples. In ICLR, 2018 a

  12. [20]

    A simple unified framework for detecting out-of-distribution samples and adversarial attacks

    Lee, K., Lee, K., Lee, H., and Shin, J. A simple unified framework for detecting out-of-distribution samples and adversarial attacks. In NeurIPS, 2018 b

  13. [21]

    and Qin, Y

    Liu, L. and Qin, Y. Fast decision boundary based out-of-distribution detector. ICML, 2024

  14. [22]

    Sphereface: Deep hypersphere embedding for face recognition, 2018

    Liu, W., Wen, Y., Yu, Z., Li, M., Raj, B., and Song, L. Sphereface: Deep hypersphere embedding for face recognition, 2018. URL https://arxiv.org/abs/1704.08063

  15. [23]

    Energy-based out-of-distribution detection

    Liu, W., Wang, X., Owens, J., and Li, Y. Energy-based out-of-distribution detection. Advances in Neural Information Processing Systems, 2020

  16. [24]

    Gen: Pushing the limits of softmax-based out-of-distribution detection

    Liu, X., Lochman, Y., and Christopher, Z. Gen: Pushing the limits of softmax-based out-of-distribution detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

  17. [25]

    Swin transformer v2: Scaling up capacity and resolution

    Liu, Z., Hu, H., Lin, Y., Yao, Z., Xie, Z., Wei, Y., Ning, J., Cao, Y., Zhang, Z., Dong, L., Wei, F., and Guo, B. Swin transformer v2: Scaling up capacity and resolution. In CVPR, 2022. URL https://arxiv.org/abs/2111.09883

  18. [26]

    Delving into out-of-distribution detection with vision-language representations

    Ming, Y., Cai, Z., Gu, J., Sun, Y., Li, W., and Li, Y. Delving into out-of-distribution detection with vision-language representations. In NeurIPS, 2022. URL https://openreview.net/forum?id=KnCS9390Va

  19. [27]

    How to exploit hyperspherical embeddings for out-of-distribution detection? In The Eleventh International Conference on Learning Representations, 2023

    Ming, Y., Sun, Y., Dia, O., and Li, Y. How to exploit hyperspherical embeddings for out-of-distribution detection? In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=aEFaE0W5pAd

  20. [28]

    and Hein, M

    Mueller, M. and Hein, M. How to train your vit for ood detection, 2024. URL https://arxiv.org/abs/2405.17447

  21. [29]

    H., and Gal, Y

    Mukhoti, J., Kirsch, A., van Amersfoort, J., Torr, P. H., and Gal, Y. Deep deterministic uncertainty: A new simple baseline. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 24384--24394, June 2023

  22. [30]

    Netzer, Y., Wang, T., Coates, A., Bissacco, A., Wu, B., and Ng, A. Y. Reading digits in natural images with unsupervised feature learning. NIPS Workshop on Deep Learning and Unsupervised Feature Learning, 2011

  23. [31]

    G., and Teoh, A

    Park, J., Jung, Y. G., and Teoh, A. B. J. Nearest neighbor guidance for out-of-distribution detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 1686--1695, 2023 a

  24. [32]

    C., Yoon, J., and Jin Teoh, A

    Park, J., Long Chai, J. C., Yoon, J., and Jin Teoh, A. B. Understanding the Feature Norm for Out-of-Distribution Detection . In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), 2023 b . URL https://doi.ieeecomputersociety.org/10.1109/ICCV51070.2023.00150

  25. [33]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., and Sutskever, I

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., and Sutskever, I. Learning transferable visual models from natural language supervision. In ICML, 2021

  26. [34]

    K., Stoyanov, D., and Bhattarai, B

    Regmi, S., Panthi, B., Dotel, S., Gyawali, P. K., Stoyanov, D., and Bhattarai, B. T2fnorm: Train-time feature normalization for ood detection in image classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pp.\ 15...

  27. [35]

    G., Padhy, S., and Lakshminarayanan, B

    Ren, J., Fort, S., Liu, J., Roy, A. G., Padhy, S., and Lakshminarayanan, B. A simple fix to mahalanobis distance for improving near-ood detection, 2021. URL https://arxiv.org/abs/2106.09022

  28. [36]

    Spreading vectors for similarity search

    Sablayrolles, A., Douze, M., Schmid, C., and J \'e gou, H. Spreading vectors for similarity search. arXiv: Machine Learning, 2018. URL https://api.semanticscholar.org/CorpusID:62841605

  29. [37]

    Ssd: A unified framework for self-supervised outlier detection

    Sehwag, V., Chiang, M., and Mittal, P. Ssd: A unified framework for self-supervised outlier detection. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=v5gjXpmR8J

  30. [38]

    P., Kolesnikov, A., Zhai, X., Wightman, R., Uszkoreit, J., and Beyer, L

    Steiner, A. P., Kolesnikov, A., Zhai, X., Wightman, R., Uszkoreit, J., and Beyer, L. How to train your vit? data, augmentation, and regularization in vision transformers. TMLR, 2022. URL https://openreview.net/forum?id=4nPswr1KcP

  31. [39]

    React: Out-of-distribution detection with rectified activations

    Sun, Y., Guo, C., and Li, Y. React: Out-of-distribution detection with rectified activations. NeurIPS, 2021

  32. [40]

    Out-of-distribution detection with deep nearest neighbors

    Sun, Y., Ming, Y., Zhu, X., and Li, Y. Out-of-distribution detection with deep nearest neighbors. ICML, 2022

  33. [41]

    Csi: Novelty detection via contrastive learning on distributionally shifted instances

    Tack, J., Mo, S., Jeong, J., and Shin, J. Csi: Novelty detection via contrastive learning on distributionally shifted instances. In NeurIPS, 2020

  34. [42]

    Hyperparameter-free out-of-distribution detection using cosine similarity

    Techapanurak, E., Suganuma, M., and Okatani, T. Hyperparameter-free out-of-distribution detection using cosine similarity. In Proceedings of the Asian Conference on Computer Vision, 2020

  35. [43]

    Deit iii: Revenge of the vit

    Touvron, H., Cord, M., and Jegou, H. Deit iii: Revenge of the vit. ECCV, 2022

  36. [44]

    The inaturalist species classification and detection dataset

    Van Horn, G., Mac Aodha, O., Song, Y., Cui, Y., Sun, C., Shepard, A., Adam, H., Perona, P., and Belongie, S. The inaturalist species classification and detection dataset. In CVPR, 2018

  37. [45]

    Open-set recognition: a good closed-set classifier is all you need? In International Conference on Learning Representations, 2022

    Vaze, S., Han, K., Vedaldi, A., and Zisserman, A. Open-set recognition: a good closed-set classifier is all you need? In International Conference on Learning Representations, 2022

  38. [46]

    Vim: Out-of-distribution with virtual-logit matching

    Wang, H., Li, Z., Feng, L., and Zhang, W. Vim: Out-of-distribution with virtual-logit matching. In CVPR, 2022

  39. [47]

    Mitigating neural network overconfidence with logit normalization

    Wei, H., Xie, R., Cheng, H., Feng, L., An, B., and Li, Y. Mitigating neural network overconfidence with logit normalization. 2022

  40. [48]

    Pytorch image models

    Wightman, R. Pytorch image models. https://github.com/rwightman/pytorch-image-models, 2019

  41. [49]

    Wilk, M. B. and Gnanadesikan, R. Probability plotting methods for the analysis for the analysis of data. Biometrika, 55 0 (1): 0 1--17, 03 1968. ISSN 0006-3444. doi:10.1093/biomet/55.1.1. URL https://doi.org/10.1093/biomet/55.1.1

  42. [50]

    S., and Xie, S

    Woo, S., Debnath, S., Hu, R., Chen, X., Liu, Z., Kweon, I. S., and Xie, S. Convnext v2: Co-designing and scaling convnets with masked autoencoders. arXiv preprint arXiv:2301.00808, 2023

  43. [51]

    Openood: Benchmarking generalized out-of-distribution detection

    Yang, J., Wang, P., Zou, D., Zhou, Z., Ding, K., Peng, W., Wang, H., Chen, G., Li, B., Sun, Y., et al. Openood: Benchmarking generalized out-of-distribution detection. arXiv preprint arXiv:2210.07242, 2022

  44. [52]

    Neural collapse with normalized features: A geometric analysis over the riemannian manifold

    Yaras, C., Wang, P., Zhu, Z., Balzano, L., and Qu, Q. Neural collapse with normalized features: A geometric analysis over the riemannian manifold. In Oh, A. H., Agarwal, A., Belgrave, D., and Cho, K. (eds.), Advances in Neural Information Processing Systems, 2022. URL https://...

  45. [53]

    Yu, C., Zhu, X., Lei, Z., and Li, S. Z. Out-of-distribution detection for reliable face recognition. IEEE Signal Processing Letters, 27: 0 710--714, 2020. doi:10.1109/LSP.2020.2988140

  46. [54]

    Anomaly detection for high-dimensional space using deep hypersphere fused with probability approach

    Zheng, J., Li, J., Liu, C., Wang, J., Li, J., and Liu, H. Anomaly detection for high-dimensional space using deep hypersphere fused with probability approach. Complex & Intelligent Systems, 8 0 (5): 0 4205--4220, Oct 2022. ISSN 2198-6053. doi:10.1007/s40747-022-00695-9. URL ht...

  47. [55]

    Places: A 10 million image database for scene recognition

    Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., and Torralba, A. Places: A 10 million image database for scene recognition. IEEE transactions on pattern analysis and machine intelligence, 40 0 (6): 0 1452--1464, 2017

  48. [56]

    M., and Ou, W

    Zhou, C., Po, L. M., and Ou, W. Angular deep supervised vector quantization for image retrieval. IEEE Transactions on Neural Networks and Learning Systems, 33 0 (4): 0 1638--1649, 2022. doi:10.1109/TNNLS.2020.3043103

Pith tools

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