Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

SimLabel: Consistency-Guided OOD Detection with Pretrained Vision-Language Models

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that scoring an image against a label's semantically related neighbors, not just the label itself, sharpens the line between in-distribution and out-of-distribution images for pretrained vision-language models.

desk verdict New consistency-over-similar-labels OOD score, but an unspecified data split in Algorithm 1 leaves the main empirical claim open to test-set leakage. read the letter →

arxiv 2501.11485 v1 pith:XNZUGZ3U submitted 2025-01-20 cs.CV

Shu Zou , Xinyu Tian , Qinyu Zhao , Zhaoyuan Yang , Jing Zhang This is my paper · ORCID
classification cs.CV
keywords out-of-distributiondetectionvision-languagemodelsCLIPzero-shotOODpost-hocscoringsemanticlabelconsistencysimilarclassgenerationimage-textalignment
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

The paper argues that pretrained vision-language models such as CLIP separate in-distribution (ID) from out-of-distribution (OOD) images more reliably when the detection score checks consistency across semantically related class labels, not just the single best-matching label. Its observation is that an ID image typically shows high similarity to a whole cluster of adjacent labels, whereas an OOD image that happens to win one label does not show that cluster-wide agreement. The paper builds a post-hoc score, SimLabel, that adds to each class affinity the average similarity over a pool of similar classes, and reports that the best variant (pools chosen by image-text alignment) reaches 91.90 average AUROC and 36.46 FPR@95 on the ImageNet-1k benchmark, against 90.62 and 43.61 for the MCM baseline, with smaller gains on fine-grained and hard-OOD tasks. The paper's own limitation analysis (Sec. VI-D) concedes that the method assumes a fairly balanced label space and weights every similar class equally, leaving long-tailed distributions and uneven semantic distances as open problems. If the claim holds, the payoff is better zero-shot OOD detection with zero training, since the new score is a drop-in replacement for the standard MCM score.

What carries the argument

The carrying mechanism is the consistency-adjusted affinity $A(x, l_c) = M(x, l_c) + \alpha \cdot \frac{1}{|D(l_c)|}\sum_{d \in D(l_c)} M(x, d)$, combined with the softmax-maximum OOD score $S(x; L, \tau) = \max_{l_c \in L} \frac{e^{A(x,l_c)/\tau}}{\sum_c e^{A(x,l_c)/\tau}}$. $M$ is the cosine similarity between a CLIP image embedding and a prompted class-label text embedding; $D(l_c)$ is that class's pool of similar labels; $\alpha$ weights the neighborhood term. The pools are the second half of the machinery: text-hierarchy selection, LLM prompting for visually similar labels, and an image-text-alignment routine (Algorithm 1) that collects, over images pseudo-classified as $l_c$, the labels that appear most often among the top-$k$ matches and keeps the top-occurring ones. The argument is that this construction converts a one-shot maximum-similarity signal into a consistency check that in-distribution images pass across the whole neighborhood and OOD images fail.

What would settle it

Run SimLabel-I twice on the ImageNet-1k benchmark, once building each class's similar-class pool exactly as Algorithm 1 specifies and once building the pools only from a training or validation subset of the in-distribution images, holding $\alpha$, $k$, and $\tau$ fixed; if average AUROC/FPR@95 falls substantially under the disjoint protocol, test-image overlap rather than label consistency is carrying the reported gain, while comparable numbers would confirm the consistency hypothesis.

Watch

Extended reading notes

Core claim

SimLabel's central claim is that the maximum-softmax score of a frozen vision-language model throws away the semantic structure of the label space. For every class $l_c$ the method builds a pool $D(l_c)$ of similar classes and defines an augmented affinity $A(x, l_c) = M(x, l_c) + \alpha \cdot \frac{1}{|D(l_c)|}\sum_{d \in D(l_c)} M(x, d)$, where $M$ is the CLIP cosine similarity between image and prompted-label embeddings; the OOD score is the maximum over classes of the softmax of these affinities at temperature $\tau$ (Eqs. 4-5). ID images are expected to score high across their class's pool, so the mean-over-neighbors term reinforces them, while an OOD image that wins a single label receives reinforcement only if its appearance happens to match the whole neighborhood, which the paper's experiments indicate it usually does not. Three pool-construction routes are tested: label hierarchy (SimLabel-H), LLM-generated visually similar classes (SimLabel-L), and counting, over ID images pseudo-labeled as $l_c$, which labels most often recur among their top-$k$ similarities (SimLabel-I), with the last performing best on the ImageNet-1k benchmark. The paper also reports that the same affinity improves zero-shot classification accuracy and transfers to other VLM backbones, and its stated scope is limited to relatively balanced label sets with equally weighted similar classes.

Load-bearing premise

The load-bearing premise is that the in-distribution images used by the pool-construction routine (Algorithm 1, Sec. IV-C3) are not the same images whose out-of-distribution scores are later computed; the paper never states whether $X_{ID}$ is training or evaluation data, and if pools are built on the evaluation images, the reported margins could rest on that overlap rather than on label-consistency.

Editorial extensions

If this is right

  • Any deployment already using the MCM score can swap in SimLabel with no retraining or OOD exemplars, because it is a post-hoc score over the same frozen image and text encoders.
  • The best variant (SimLabel-I) also improves zero-shot classification accuracy on ImageNet, ImageNetV2, and CUB-200, so the neighborhood term helps in-distribution recognition rather than only OOD separation.
  • The reported gains transfer across model families (CLIP, AltCLIP, GroupViT), across ID scales from ImageNet-1k to fine-grained sets, and to semantic-hard splits, which the paper offers as evidence of generality.
  • A similar-classes-only score (SimLabel-S) that drops the ground-truth label still detects OOD inputs far better than chance, supporting the paper's claim that adjacent labels share prototype structure.
  • A small pool of $k = 6$ similar classes per class suffices; the paper reports diminishing returns as $k$ grows, which justifies its hyper-parameter choice.

Reading between the lines

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

  • Editorial check: if Algorithm 1's $X_{ID}$ is the evaluation set, the reported margins over MCM are not a clean test of the consistency idea; a strictly disjoint pool (training or validation images only) is the decisive experiment, and the released code can settle which split is used.
  • The paper's pooled mean over similar classes is one summary of the top-$k$ similarity distribution; untested alternatives such as the variance or entropy of an image's similarities across its predicted class's neighborhood would probe the same consistency hypothesis without needing an explicit label pool.
  • A natural downstream application the paper does not evaluate is prediction refusal: if a class wins without neighborhood support, the same affinity could flag the input as low-confidence for abstention in zero-shot classifiers.
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

4 major / 6 minor

Summary. The paper proposes SimLabel, a post-hoc zero-shot OOD detection score for pretrained vision-language models. For each in-distribution class label, SimLabel constructs a set of semantically similar labels using one of three strategies: label-set hierarchy, LLM-generated world knowledge, or image-text alignment on in-distribution images. The final affinity in Eq. (4) averages the CLIP similarity over these similar labels and adds it to the class-label similarity, and the score in Eq. (5) is the maximum softmax value of the resulting affinities. Experiments on the ImageNet-1k benchmark and on fine-grained datasets report improvements over the MCM baseline and several other post-hoc methods, with additional experiments on hard OOD tasks, other VLM backbones, and classification accuracy.

Significance. If the reported gains are robust and leakage-free, SimLabel is a simple and useful contribution: it shows that aggregating similarity over semantically related labels can improve ID/OOD separability without training any parameters. The paper is generally clearly written, the formulation is easy to reproduce from Eqs. (4) and (5), and the authors include useful ablations, including a version that uses only similar-class affinities (SimLabel-S) and zero-shot classification results. The main significance is currently limited by the modest and dataset-dependent gains, the lack of error bars, and a load-bearing ambiguity about whether Algorithm 1 uses the evaluation split to construct the similar-class pool.

major comments (4)
  1. [Sec. IV-C3, Algorithm 1] Algorithm 1 does not specify what XID is. If XID is the same ImageNet validation split used to compute the AUROC/FPR95 numbers in Table II, then the similar-class sets D(lc) are constructed by pseudo-labeling and top-k selection on the very images that are later scored. This makes the procedure transductive rather than the claimed zero-shot post-hoc method, and it can inflate the reported improvements independently of the consistency hypothesis. The authors should state the split explicitly and, ideally, construct D(lc) using a disjoint set of ID images and re-report Tables II, III, and IV under that protocol.
  2. [Secs. V-A and V-C, Table VI and Fig. 5] The hyperparameters alpha=1 and k=6 are selected using the same ImageNet-1k benchmark on which the main results are reported, with no separate validation split described. Table VI and Fig. 5 are sensitivity analyses on the evaluation benchmark itself, so the reported numbers are selection results rather than an unbiased estimate of expected performance. Please fix hyperparameters on a held-out validation split or, at minimum, report the sensitivity of all reported datasets to alpha and k and justify the final choice under a predefined protocol.
  3. [Table II] The claimed improvement over the strongest baselines is not uniform across datasets. For example, SimLabel-I has lower AUROC than NPOS on SUN (90.35 vs. 90.44) and Textures (87.07 vs. 88.80), and substantially worse FPR95 on Textures (53.65 vs. 46.12). The paper averages over four OOD datasets and does not report variance or significance tests, which makes it difficult to tell whether the consistency mechanism, rather than dataset-specific label choices, drives the average gain. Please add per-dataset confidence intervals or repeated-run statistics and discuss the conditions under which the consistency score is expected to help.
  4. [Sec. IV-C3 and Eq. (4)] The paper does not include a controlled comparison that isolates the effect of label-set construction from the consistency mechanism. Since SimLabel-I obtains D(lc) from image-text alignment on XID, its label pool is potentially tailored to the test distribution, whereas SimLabel-H and SimLabel-L use external text information. A control using random label sets of the same cardinality, or using hierarchy-derived sets matched for size, under the same data-split conditions, would directly test whether the observed gains are due to consistency over similar labels or to the information used to select them.
minor comments (6)
  1. [Sec. III, Eq. (1)] The function domains in Eq. (1) are swapped: f_T should map text prompts to R^D and f_I should map images to R^D as written in the text, and Eq. (2) has an unmatched closing parenthesis in the cosine-similarity expression.
  2. [Sec. IV-C3] The phrase "set of similar class (donates D(xc))" should read "denotes D(xc)", and the notation D(xc) is overloaded: it is used both for the per-image set of top-k labels and for the aggregated class-wise set D(lc). A clearer notation would avoid ambiguity.
  3. [Table VI] Table VI reports FPR95 values but the metric is identified only in the surrounding text; adding a row or column label such as "FPR95" inside the table would make it self-contained.
  4. [Table V caption] The caption contains a typo, "ImgeNet", and the text refers inconsistently to "ImageNet-100"; please unify the dataset name and spelling.
  5. [Sec. II] The description of MCM as using "the maximum similarity" is imprecise because MCM applies a softmax over the cosine similarities and uses the maximum softmax value; the paper should state this consistently with Sec. IV-A and Fig. 2.
  6. [Sec. II and References] The reference [14] is described as both "CLIPEN" and "Dai et al."; please unify the citation name to avoid reader confusion.

Circularity Check

2 steps flagged · score 6.0 of 10

SimLabel-I's similar-class pool is built from the same XID images that are later scored, making the consistency term a self-aggregated summary of the scored similarities.

  1. other [Algorithm 1 (Sec. IV-C3) and Eq. 4-5 in Sec. IV-D]
    "Require: ID label set L, ID sample xID ∈ XID ... Xc ⊂ XID // Subset of XID with Eq. 2 whose pseudo label is lc ... D(lc) ← Select(D(Xc)) // Select labels in D(xc) with top-k highest occurrence. / A(x, lc) =M(x, lc) +α ∗ X d∈D(lc) M(x, d)/|D(lc)|"

    The paper labels SimLabel a zero-shot, post-hoc method and the ImageNet-1k benchmark uses ImageNet-1k as ID data, so XID in Algorithm 1 is the same set of images whose ID/OOD scores are computed by Eq. 5. D(lc) is chosen because images x in Xc, including the scored ID images, had top-k similarity to those labels. Eq. 4 then averages M(x,d) over exactly those self-selected labels. The consistency contribution to S(x) is therefore a function of x's own top-k similarities rather than an independent consistency measurement; the reported gains over MCM can be inflated by this transductive self-scoring even if the semantic-consistency hypothesis were false.

  2. self definitional [Sec. VI-B, reformulated isolated affinity]
    "To verify this, we reformulate the image-class prototype similarity defined in Eq. 5 to consider only the similarity between the image and its similar-class labels. ... A(x, lc) = Σ_{d∈D(lc)} M(x,d)/|D(lc)|. ... we have applied the affinity metric for OOD detection using the similar classes generated for SimLabel-I on the ImageNet benchmark."

    This is presented as verification of the assumption that ID images show consistent high similarity to their similar classes, but D(lc) was generated by Algorithm 1 from the same ImageNet ID images that are subsequently scored. The 'isolated' affinity is not an independent test: positive SimLabel-S results are produced by reusing the same top-k occurrence selection criterion on the same images, so the assumption is confirmed by construction rather than by external evidence.

full rationale

The derivation of the SimLabel score itself is transparent and the softmax structure in Eq. 5 is a straightforward extension of MCM, so there is no hidden algebraic equivalence to the inputs. However, the main variant SimLabel-I has a partially circular validation path. Algorithm 1 obtains the similar-class pool D(lc) from XID without stating a held-out split; in the zero-shot ImageNet-1k protocol, XID is the ID evaluation set. Each scored ID image therefore contributes to selecting the labels whose similarities are later averaged into its own score, so the consistency signal is partly self-confirming. The Sec. VI-B 'verification' of the consistency assumption compounds this by testing on the same constructed pool. SimLabel-H and SimLabel-L rely on external hierarchy or LLM knowledge and are not subject to this particular circularity, but the paper's headline result and the focus of the empirical analysis is SimLabel-I. The hyper-parameter choices (alpha=1, k=6) are tuned on the benchmark and reported as fixed values, which is a standard but additional source of optimistic evaluation; it is not itself circular. Overall, the central empirical claim is partially reduced to a transductive construction from the scored inputs, so a score of 6 is warranted.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical or mathematical entities. It relies on three free parameters (alpha, k, tau), an unverified data-split assumption in Algorithm 1, and a domain assumption about ID similarity consistency.

free parameters (3)
  • alpha = 1.0
    Weight of the similar-class affinity in Eq. 4, tuned by ablations on the ImageNet-1k benchmark (Table VI).
  • k = 6
    Number of similar classes per class, selected based on the FPR95 ablation in Fig. 5.
  • temperature tau = 1.0
    Softmax temperature in Eq. 5, taken directly from MCM [13], not tuned here but the central score depends on it.
assumptions (4)
  • domain assumption ID samples should consistently show high similarity to a set of semantically related ID classes, not just to their ground-truth class.
    This is the core hypothesis motivating SimLabel, stated in Sec. IV-B and used in the score definition (Eqs. 4-5).
  • domain assumption CLIP pseudo-labels used in Algorithm 1 are accurate enough that the selected similar classes represent the class prototype.
    Algorithm 1 relies on pseudo-labeling of ID images to aggregate similar labels per class.
  • domain assumption All similar classes contribute equally to the affinity in Eq. 4.
    The paper assumes equal weighting, later acknowledging this as a limitation in Sec. VI.D.
  • standard math Cosine similarity and softmax are valid measures for OOD scoring.
    The score construction follows the standard MCM framework and common practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SimLabel: Consistency-Guided OOD Detection with Pretrained Vision-Language Models." pith.science (2026). https://pith.science/paper/XNZUGZ3U

@misc{pith2026250111485,
  author       = {Pith},
  title        = {Pith review of: SimLabel: Consistency-Guided OOD Detection with Pretrained Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XNZUGZ3U}},
  note         = {Machine review of arXiv:2501.11485}
}
read the original abstract

Detecting out-of-distribution (OOD) data is crucial in real-world machine learning applications, particularly in safety-critical domains. Existing methods often leverage language information from vision-language models (VLMs) to enhance OOD detection by improving confidence estimation through rich class-wise text information. However, when building OOD detection score upon on in-distribution (ID) text-image affinity, existing works either focus on each ID class or whole ID label sets, overlooking inherent ID classes' connection. We find that the semantic information across different ID classes is beneficial for effective OOD detection. We thus investigate the ability of image-text comprehension among different semantic-related ID labels in VLMs and propose a novel post-hoc strategy called SimLabel. SimLabel enhances the separability between ID and OOD samples by establishing a more robust image-class similarity metric that considers consistency over a set of similar class labels. Extensive experiments demonstrate the superior performance of SimLabel on various zero-shot OOD detection benchmarks. The proposed model is also extended to various VLM-backbones, demonstrating its good generalization ability. Our demonstration and implementation codes are available at: https://github.com/ShuZou-1/SimLabel.

Figures

Figures reproduced from arXiv: 2501.11485 by the authors.

Figure 1
Figure 1. (a) Illustration of VLMs guided OOD detection for ID (top image from ImageNet [ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. This figure demonstrate the sorted average similarity between a specific [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. This figure illustrates samples of similar classes for the class ”Great White Shark” using methods in Sec. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Overview of the SimLabel zero-shot OOD detection framework. The image encoder first encodes ID and OOD images into image embeddings h and h ′ , respectively. For every class label (represented as blue blocks) in the ID label set L, similar classes (represented as yello…
Figure 5
Figure 5. Figure 5: This figure illustrates how the FPR@95 changes with different choices on the number of similar classes ( [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: This figure presents two practical examples illustrating how our [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID 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. High-Entropy Tokens as Multimodal Failure Points in Vision-Language Models

    cs.CV 2025-12 unverdicted novelty 6.0 of 10

    High-entropy tokens act as concentrated multimodal failure points in VLMs, enabling sparse Entropy-Guided Attacks that achieve 93-95% success and 30-38% harmful rates with cross-model transfer.

Reference graph

Works this paper leans on

50 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    A baseline for detecting misclassified and out-of-distribution examples in neural networks,

    D. Hendrycks and K. Gimpel, “A baseline for detecting misclassified and out-of-distribution examples in neural networks,” arXiv preprint arXiv:1610.02136, 2016

    arXiv 2016

  2. [2]

    Predictive sample as- signment for semantically coherent out-of-distribution detection,

    Z. Peng, E. Wang, X. Liu, and M.-M. Cheng, “Predictive sample as- signment for semantically coherent out-of-distribution detection,” IEEE Transactions on Circuits and Systems for Video Technology , pp. 1–1, 2024

    work page 2024

  3. [3]

    Openmix+: Revisiting data augmentation for open set recognition,

    G. Jiang, P. Zhu, Y . Wang, and Q. Hu, “Openmix+: Revisiting data augmentation for open set recognition,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 33, no. 11, pp. 6777–6787, 2023

    work page 2023

  4. [4]

    Pretrained transformers improve out-of-distribution robust- ness,

    D. Hendrycks, X. Liu, E. Wallace, A. Dziedzic, R. Krishnan, and D. Song, “Pretrained transformers improve out-of-distribution robust- ness,” arXiv preprint arXiv:2004.06100 , 2020

    arXiv 2004

  5. [5]

    Generalized odin: Detecting out-of-distribution image without learning from out-of-distribution data,

    Y .-C. Hsu, Y . Shen, H. Jin, and Z. Kira, “Generalized odin: Detecting out-of-distribution image without learning from out-of-distribution data,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10951–10960, 2020

    work page 2020

  6. [6]

    Towards textual out-of-domain detection without in-domain labels,

    D. Jin, S. Gao, S. Kim, Y . Liu, and D. Hakkani-T ¨ur, “Towards textual out-of-domain detection without in-domain labels,” IEEE/ACM Trans- actions on Audio, Speech, and Language Processing , vol. 30, pp. 1386– 1395, 2022

    work page 2022

  7. [7]

    Enhancing the generalization for intent classification and out-of-domain detection in slu,

    Y . Shen, Y .-C. Hsu, A. Ray, and H. Jin, “Enhancing the generalization for intent classification and out-of-domain detection in slu,” arXiv preprint arXiv:2106.14464, 2021

    arXiv 2021

  8. [8]

    Unsupervised out-of-domain detection via pre-trained transformers,

    K. Xu, T. Ren, S. Zhang, Y . Feng, and C. Xiong, “Unsupervised out-of-domain detection via pre-trained transformers,” arXiv preprint arXiv:2106.00948, 2021

    arXiv 2021

Show all 50 references
  1. [9]

    Classifier-head informed feature masking and prototype-based logit smoothing for out- of-distribution detection,

    Z. Sun, Y . Qiu, Z. Tan, W. Zheng, and R. Wang, “Classifier-head informed feature masking and prototype-based logit smoothing for out- of-distribution detection,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 7, pp. 5630–5640, 2024

    2024

  2. [10]

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

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

    2022

  3. [11]

    Learning transferable visual models from natural language supervi- sion,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” 2021

    2021

  4. [12]

    Zero-shot out-of- distribution detection based on the pre-trained model clip,

    S. Esmaeilpour, B. Liu, E. Robertson, and L. Shu, “Zero-shot out-of- distribution detection based on the pre-trained model clip,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 36, p. 6568–6576, June 2022

    2022

  5. [13]

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

    Y . Ming, Z. Cai, J. Gu, Y . Sun, W. Li, and Y . Li, “Delving into out-of- distribution detection with vision-language representations,” 2022

    2022

  6. [14]

    Exploring large language models for multi-modal out-of-distribution detection,

    Y . Dai, H. Lang, K. Zeng, F. Huang, and Y . Li, “Exploring large language models for multi-modal out-of-distribution detection,” 2023

    2023

  7. [15]

    Clipn for zero-shot ood detection: Teaching clip to say no,

    H. Wang, Y . Li, H. Yao, and X. Li, “Clipn for zero-shot ood detection: Teaching clip to say no,” 2023

    2023

  8. [16]

    Nega- tive label guided ood detection with pretrained vision-language models,

    X. Jiang, F. Liu, Z. Fang, H. Chen, T. Liu, F. Zheng, and B. Han, “Nega- tive label guided ood detection with pretrained vision-language models,” in International Conference on Learning Representations (ICLR) , 2024

    2024

  9. [17]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in 2009 IEEE Conference on Computer Vision and Pattern Recognition , pp. 248–255, 2009

    2009

  10. [18]

    The inaturalist species classification and detection dataset,

    G. V . Horn, O. M. Aodha, Y . Song, Y . Cui, C. Sun, A. Shepard, H. Adam, P. Perona, and S. Belongie, “The inaturalist species classification and detection dataset,” 2018

    2018

  11. [19]

    Enhancing the reliability of out-of- distribution image detection in neural networks,

    S. Liang, Y . Li, and R. Srikant, “Enhancing the reliability of out-of- distribution image detection in neural networks,” 2020

    2020

  12. [20]

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

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

    2022

  13. [21]

    Energy-based out-of- distribution detection,

    W. Liu, X. Wang, J. D. Owens, and Y . Li, “Energy-based out-of- distribution detection,” 2021

    2021

  14. [22]

    React: Out-of-distribution detection with rectified activations,

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

    2021

  15. [23]

    Dice: Leveraging sparsification for out-of-distribution detection,

    Y . Sun and Y . Li, “Dice: Leveraging sparsification for out-of-distribution detection,” 2022

    2022

  16. [24]

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

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

    2018

  17. [25]

    Out-of-distribution detection with deep nearest neighbors,

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

    2022

  18. [26]

    Exploring the limits of out- of-distribution detection,

    S. Fort, J. Ren, and B. Lakshminarayanan, “Exploring the limits of out- of-distribution detection,” 2021

    2021

  19. [27]

    Non-parametric outlier synthesis,

    L. Tao, X. Du, X. Zhu, and Y . Li, “Non-parametric outlier synthesis,” 2023

    2023

  20. [28]

    Improving zero-shot generalization and robustness of multi-modal models,

    Y . Ge, J. Ren, A. Gallagher, Y . Wang, M.-H. Yang, H. Adam, L. Itti, B. Lakshminarayanan, and J. Zhao, “Improving zero-shot generalization and robustness of multi-modal models,” 2023

    2023

  21. [29]

    Are we done with imagenet?,

    L. Beyer, O. J. H ´enaff, A. Kolesnikov, X. Zhai, and A. v. d. Oord, “Are we done with imagenet?,” arXiv preprint arXiv:2006.07159 , 2020

    2006 arXiv

  22. [30]

    Self-consistency improves chain of thought reasoning in language models,

    X. Wang, J. Wei, D. Schuurmans, Q. Le, E. Chi, S. Narang, A. Chowdh- ery, and D. Zhou, “Self-consistency improves chain of thought reasoning in language models,” 2023

    2023

  23. [31]

    Chils: Zero-shot image classification with hierarchical label sets,

    Z. Novack, J. McAuley, Z. C. Lipton, and S. Garg, “Chils: Zero-shot image classification with hierarchical label sets,” 2023

    2023

  24. [32]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” Advances in neural informa- tion processing systems , vol. 25, 2012

    2012

  25. [33]

    Do convolutional neural networks learn class hierarchy?,

    A. Bilal, A. Jourabloo, M. Ye, X. Liu, and L. Ren, “Do convolutional neural networks learn class hierarchy?,” IEEE transactions on visualiza- tion and computer graphics , vol. 24, no. 1, pp. 152–162, 2017

    2017

  26. [34]

    Hiernet: Image recognition with hierarchical convolutional networks.,

    L. Tempfli and C. S ´andor, “Hiernet: Image recognition with hierarchical convolutional networks.,” in ICAART (2), pp. 147–155, 2024

    2024

  27. [35]

    Language models are few-shot learners,

    T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert- V oss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D. M. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, S. Gray, B...

    2020

  28. [36]

    Zero-shot in-distribution detection in multi-object settings using vision-language foundation mod- els,

    A. Miyai, Q. Yu, G. Irie, and K. Aizawa, “Zero-shot in-distribution detection in multi-object settings using vision-language foundation mod- els,” 2023

    2023

  29. [37]

    Hard but robust, easy but sensitive: How encoder and decoder perform in neural machine translation,

    T. He, X. Tan, and T. Qin, “Hard but robust, easy but sensitive: How encoder and decoder perform in neural machine translation,” arXiv preprint arXiv:1908.06259, 2019

    1908 arXiv

  30. [38]

    On the importance of gradients for detecting distributional shifts in the wild,

    R. Huang, A. Geng, and Y . Li, “On the importance of gradients for detecting distributional shifts in the wild,” 2021

    2021

  31. [39]

    Sun database: Large-scale scene recognition from abbey to zoo,

    J. Xiao, J. Hays, K. A. Ehinger, A. Oliva, and A. Torralba, “Sun database: Large-scale scene recognition from abbey to zoo,” in 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 3485–3492, 2010

    2010

  32. [40]

    Places: An image database for deep scene understanding,

    B. Zhou, A. Khosla, A. Lapedriza, A. Torralba, and A. Oliva, “Places: An image database for deep scene understanding,” 2016

    2016

  33. [41]

    Describing textures in the wild,

    M. Cimpoi, S. Maji, I. Kokkinos, S. Mohamed, and A. Vedaldi, “Describing textures in the wild,” 2013

    2013

  34. [42]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proceedings of the 31st International Conference on Neural Information Processing Systems , NIPS’17, (Red Hook, NY , USA), p. 6000–6010, Curran...

    2017

  35. [43]

    Gpt-4 technical report,

    OpenAI, J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, and S. A. et al, “Gpt-4 technical report,” 2024

    2024

  36. [44]

    Caltech-ucsd birds 200,

    P. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Belongie, and P. Perona, “Caltech-ucsd birds 200,” 09 2010

    2010

  37. [45]

    Food-101 – mining dis- criminative components with random forests,

    L. Bossard, M. Guillaumin, and L. Van Gool, “Food-101 – mining dis- criminative components with random forests,” in European Conference on Computer Vision , 2014

    2014

  38. [46]

    Cats and dogs,

    O. M. Parkhi, A. Vedaldi, A. Zisserman, and C. V . Jawahar, “Cats and dogs,” in 2012 IEEE Conference on Computer Vision and Pattern Recognition, pp. 3498–3505, 2012

    2012

  39. [47]

    3d object representations for fine-grained categorization,

    J. Krause, M. Stark, J. Deng, and L. Fei-Fei, “3d object representations for fine-grained categorization,” in 2013 IEEE International Conference on Computer Vision Workshops , pp. 554–561, 2013

    2013

  40. [48]

    Altclip: Al- tering the language encoder in clip for extended language capabilities,

    Z. Chen, G. Liu, B.-W. Zhang, F. Ye, Q. Yang, and L. Wu, “Altclip: Al- tering the language encoder in clip for extended language capabilities,” arXiv preprint arXiv:2211.06679 , 2022

    2022 arXiv

  41. [49]

    Groupvit: Semantic segmentation emerges from text supervision,

    J. Xu, S. De Mello, S. Liu, W. Byeon, T. Breuel, J. Kautz, and X. Wang, “Groupvit: Semantic segmentation emerges from text supervision,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18134–18144, 2022

    2022

  42. [50]

    Do imagenet classifiers generalize to imagenet?,

    B. Recht, R. Roelofs, L. Schmidt, and V . Shankar, “Do imagenet classifiers generalize to imagenet?,” 2019

    2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.