Pith. sign in

REVIEW 4 major objections 5 minor 31 references

Dual Anchors, Do It Better: Hierarchical Group Merging for Zero-Shot Anomaly Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read The paper proposes a Dual-Anchor framework that adds hierarchical image anchors to text prompts and reports state-of-the-art zero-shot anomaly detection across 14 benchmarks.

desk verdict The dual-anchor idea is plausible and worth watching, but the forward pass as written is dimensionally impossible—1024-d image tokens fed into a 768-d text encoder with no projection—so the reported numbers are unsupported until that is fixed. read the letter →

arxiv 2608.11933 v1 pith:CJAYLOI3 submitted 2026-08-12 cs.CV

Jimin Roh , DongKyu Kim , Suk-Ju Kang This is my paper · ORCID
classification cs.CV
keywords zero-shotanomalydetectiondual-anchorframeworkhierarchicalgroupmergingdynamicstatepromptCLIPDINOv3localizationindustrialandmedicalbenchmarks
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish a new recipe for zero-shot anomaly detection—spotting defects in domains never seen during training: instead of anchoring normal and abnormal semantics only in text prompts, it builds visual anchors by hierarchically merging image features, then fuses both kinds of anchors. The proposed Dual-Anchor framework forms normal and abnormal group tokens from DINOv3 visual features, uses those tokens to gate and refine the global image token, and inserts them into CLIP prompts so the text branch becomes image-dependent. Across 8 industrial and 6 medical benchmarks, the paper reports the best image-level and pixel-level results among the compared zero-shot anomaly detection methods, including gains over the strongest existing baseline. If the claim is right, prompt sensitivity and text bias are not unavoidable weaknesses of CLIP-based anomaly detection, because structured visual grouping can carry part of the semantic load.

What carries the argument

The load-bearing objects are the two learnable image anchors $G_N$ and $G_A$ produced by Hierarchical Group Merging. Each Merge Block runs four stages—initialization by cross-attention with previous group tokens, Gumbel-Softmax assignment of patch tokens to groups, bipartite soft matching that merges the most similar group pairs, and a final update that assigns merged tokens back to the patches—so the two anchors accumulate local-to-global visual semantics. The Group-Gated Token Refiner computes cosine gating weights between the [CLS] token and the two anchors, passes the concatenated weights through normalization, a linear layer, and a sigmoid, then adds the result to [CLS] as a residual connection. The Dynamic State Prompt concatenates the visual group tokens into text prompts before CLIP encoding, which is the step that makes the text embedding image-dependent and directly reduces reliance on prompt wording.

What would settle it

Implement the dynamic state prompt exactly as written in Eq. (12): the 1024-dimensional $G_N$ token cannot pass through the 768-dimensional token embedding layer of the CLIP text encoder, so the missing projection must be specified before the reported gains can be attributed to the prompt. Then run the prompt ablation with the group tokens replaced by random 1024-dimensional vectors; if the image-level AUROC on MVTec-AD stays near the reported 92.7, the visual conditioning is not carrying the claimed signal.

Watch

Extended reading notes

Core claim

The central claim is that anchoring semantics on both modalities—text prompts plus hierarchical image anchors—yields more stable and more generalizable zero-shot anomaly detection than text-only anchoring. The paper constructs image anchors with a top-down Hierarchical Group Merging process that assigns DINOv3 patch tokens to learnable groups, merging the most similar groups until two final group tokens $G_N$ and $G_A$ stand for normal and abnormal states. These two tokens gate the [CLS] token through a Group-Gated Token Refiner and are inserted into dynamic state prompts of the form $[V_1]\cdots[V_E][W_1][G_N][\text{class}]$, making the text branch depend on the input image. The paper reports that this design outperforms all compared baselines on 8 industrial and 6 medical benchmarks, with the full model reaching 92.7 image-level AUROC and 92.4 pixel-level AUROC on MVTec-AD, and that removing any of the three modules lowers performance.

Load-bearing premise

The paper's dynamic state prompt inserts 1024-dimensional visual group tokens directly into a CLIP text encoder that embeds 768-dimensional tokens, and no projection or adapter for that dimension mismatch is described.

Editorial extensions

If this is right

  • Zero-shot anomaly detection no longer has to rely on carefully engineered or heavily tuned prompt wording; the image group tokens carry normal and abnormal semantics into the text branch.
  • Hierarchical merging of visual features provides a local-to-global representation that improves pixel-level localization, not just image-level classification, on both object-centric and texture datasets.
  • The dual-anchor design transfers across domain boundaries: the same auxiliary-trained recipe works on industrial defects and medical lesions without any target-domain labels.
  • Ablations show each component is load-bearing: removing Hierarchical Group Merging, the Group-Gated Token Refiner, or the Dynamic State Prompt each lowers both image-level and pixel-level metrics.
  • The number of group tokens per merge stage is a real hyperparameter, with [16, 8, 4, 2] outperforming both smaller and larger settings in the paper's ablation.

Reading between the lines

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

  • The dual-anchor recipe is not tied to CLIP's text encoder: the same hierarchical image anchors could be paired with other vision-language encoders, since the anchors are learned from auxiliary data rather than from target labels.
  • Because the image anchors are built without target-domain labels, the approach should extend naturally to few-shot anomaly detection, where a few normal examples could refine the group tokens before deployment.
  • A concrete stress test the paper does not run is a prompt-sensitivity sweep: varying the wording of the class prompt while keeping the visual anchors fixed would quantify how much of the reported stability comes from the image side.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a zero-shot anomaly detection (ZSAD) framework that augments the usual text-side anchors with image-side anchors. It extracts 1024-dimensional patch features from DINOv3, progressively merges them into normal/abnormal group tokens via hierarchical merge blocks, uses these tokens to refine the [CLS] token through a Group-Gated Token Refiner, and injects them into CLIP text prompts to form dynamic state prompts. Training on an auxiliary dataset uses classification, segmentation, orthogonality, and group-consistency losses. The method is evaluated on 8 industrial and 6 medical benchmark datasets, with average improvements over six prior ZSAD baselines, supported by component ablations and qualitative visualizations.

Significance. If the method can be made fully specified and reproducible, the dual-anchor idea is a plausible step toward reducing the prompt sensitivity that limits CLIP-based ZSAD. The paper's strengths include broad benchmark coverage, careful module-level ablations, and an explicit attempt to ground text semantics in hierarchical visual structure. However, the current manuscript has a load-bearing dimensional inconsistency in its cross-modal computations, and the shared use of image-derived tokens on both sides of the alignment creates a potential shortcut that is not controlled. These issues must be resolved before the empirical claims can be assessed.

major comments (4)
  1. [§3.4–§3.5, Eq. 12, Eq. 13, Eq. 15; Implementation Details] The forward pass is dimensionally inconsistent with the stated backbones. Implementation Details explicitly says DINOv3 ViT-L/16 yields 1024-dimensional patch features and CLIP ViT-L/14@336 yields 768-dimensional embeddings. Eq. 12 inserts the 1024-d group tokens G_N and G_A directly into the prompt consumed by the 768-d CLIP text encoder; Eq. 13 multiplies the 1024-d concatenated visual feature block by T^T, where T is 2 x 768; Eq. 15 computes cosine similarity between the 1024-d refined [CLS] token and the 768-d text embeddings. No projection, adapter, or shared embedding space is described, and none of the losses in Eqs. 16-19 train such a layer. As written, the method cannot execute, so the results in Tables 1-4 are not reproducible. Please specify the missing projection/adaptation, retrain if necessary, and re-report, or clearly remove the unsupported dynamic-state-prompt results.
  2. [§3.4 and §3.5, Eq. 12 with Eq. 15] The design shares the same image-derived group tokens across both branches: G_N and G_A refine the [CLS] token (Eqs. 10-11), and the same tokens are injected into the text prompts whose embeddings are then compared back to t_refined_cls (Eq. 15). This creates a possible shortcut where the normal/abnormal text embeddings are partly a function of the test image itself, so high similarity may reflect the injected visual tokens rather than genuine image-text semantic alignment. The claim that the framework 'stabilizes image-text alignment' needs a control experiment—for example, ablating with group tokens taken from a different image, or removing the prompt injection while keeping the GGTR—to show that the discriminative signal does not come primarily from the shared anchors.
  3. [§4.2, Tables 1 and 2] The sentence in §4.2 that 'our model consistently outperforms all existing baselines across both industrial and medical benchmarks' is contradicted by the tables. Bayes-PFL reports higher pixel AUROC than the proposed method on MPDD (97.1 vs 95.9), RSDD (99.6 vs 99.2), KSDD2 (99.6 vs 99.5), and DAGM (99.3 vs 98.3); AA-CLIP reports higher ISIC AP (87.7 vs 84.7); and Bayes-PFL reports higher CVC-ClinicDB AP (53.2 vs 39.8). The claim should be revised to 'on average' or supported by a significance test over datasets.
  4. [§4.3, Table 3(a)] The ablation labeled 'w/o Hierarchical Group Merging' removes only the assigned token F_assign, but the group tokens G_N and G_A that are the direct outputs of the merging hierarchy are still used by the Group-Gated Token Refiner and the Dynamic State Prompt. Therefore this ablation does not isolate the contribution of the hierarchical merging mechanism. A cleaner ablation would remove the entire group-token generation path while retaining the same number of parameters where possible, so that the role of the hierarchical grouping itself is measured.
minor comments (5)
  1. [§3.3] In the 'Update' paragraph, the text begins 'Update: Update:' with the word 'Update' duplicated; this appears to be a typo.
  2. [§3.5, Eq. 15] The denominator in Eq. 15 is typeset incorrectly; it should read the product of the L2 norms of t_refined_cls and T.
  3. [Author affiliations] The affiliation 'NA VER Cloud' appears to be a typo for 'NAVER Cloud'; please correct it.
  4. [§3.5, Eq. 13] The notation [1/L sum_l F_i^(l), F_assign] should specify whether concatenation is along the token dimension or the feature dimension; as written, the resulting matrix dimension is ambiguous even apart from the dimension mismatch with T.
  5. [§4.1, Implementation Details] Please state the exact temperature values for the Gumbel-Softmax assignment (Eq. 2) and the cosine similarity (Eq. 15), as well as the loss weighting coefficients in Eq. 19; these are omitted from the main text and the supplementary material is not available in the reviewed version.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the dual-anchor mechanism is an explicit architecture trained on auxiliary data, not a derivation that reduces to its own inputs.

full rationale

The paper's derivation chain is self-contained as an experimental method rather than a formal derivation. The group tokens G_N and G_A are produced from the input image by a frozen DINOv3 backbone plus learned hierarchical grouping, and the grouping parameters are trained on an auxiliary dataset (MVTec or VisA), not on the target-domain labels that are later predicted. The Dynamic State Prompt (Eq. 12) intentionally injects these image-derived group tokens into the text prompt, and the final anomaly scores (Eqs. 13 and 15) therefore compare image features with text embeddings that are conditioned on the same image. This is an explicit modeling choice, and it does not make the predicted anomaly score equal to the input by construction; the score is a nonlinear function of the image through two different branches (the group-gated CLS refinement and the prompt-conditioned text encoder) with learned parameters. No fitted parameter is renamed as a prediction, and no load-bearing self-citation or imported uniqueness theorem is used. The dimensional inconsistency between 1024-dimensional DINOv3 tokens and 768-dimensional CLIP embeddings is a reproducibility/correctness concern, not a circularity of the derivation chain. Overall, the central claim rests on empirical comparison against external baselines and ablations, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 1 invented entities

The central result rests on: (1) the standard ZSAD protocol of training on MVTec/VisA and testing on unseen domains; (2) an unstated projection that makes DINOv3 group tokens usable inside the CLIP text encoder (Eq. 12); (3) hyperparameters chosen via ablations on target benchmarks (Table 4); and (4) several loss weights and temperatures that are never reported. These are not independent external benchmarks, so the ledger is dominated by domain assumptions and paper-specific design choices rather than free constants fitted to physical data.

free parameters (3)
  • group token configuration per merge block = [16, 8, 4, 2] (final stage fixed at 2)
    Chosen from Table 4 ablation on target benchmarks; variant [2,2,2,2] yields image-level AUROC 88.2 vs 92.7 for the chosen setting, so the central result depends on this choice.
  • loss weights for L_cls, L_seg, L_ortho, L_group = not reported (all implied equal)
    Equation 19 sums all four losses without coefficients; the relative weighting is a free choice that is never stated.
  • temperatures tau (Gumbel Eq. 2 and similarity Eq. 15) = not reported
    tau controls assignment hardness and logit scale; values are not given in the main text.
assumptions (5)
  • standard math Gumbel-Softmax relaxation is differentiable and approximates hard assignment (Eqs. 2-3).
    Invoked in Sec. 3.3 to justify end-to-end training of discrete group assignment.
  • domain assumption DINOv3 features from layers 6, 12, 18, and 24 provide sufficient visual hierarchy for grouping normal and anomalous patches in unseen domains.
    Stated in Sec. 3.2 without evidence; the entire grouping mechanism relies on this transfer.
  • domain assumption Auxiliary training on MVTec-AD or VisA with pixel-level masks transfers to 12 other industrial and medical domains.
    This is the standard zero-shot anomaly detection protocol, and all reported results depend on it.
  • ad hoc to paper The first two group tokens after the final merge represent normal and anomalous semantics respectively (Eq. 9), enforced by L_group (Eq. 18).
    This semantic split is not guaranteed by the grouping mechanism; it requires ground-truth masks during training.
  • ad hoc to paper Top-r bipartite matching (Eqs. 5-8) preserves anomaly-relevant semantics during token reduction.
    Inherited from ToMe [4] for efficiency, but not validated for anomaly semantics in unseen domains.
invented entities (1)
  • normal and anomalous group tokens G_N and G_A (image anchors)
    purpose: Serve as visual counterparts to text prompts, gate the [CLS] token, and condition dynamic prompts.
    Learned constructs supervised by ground-truth masks through Eq. 18; no external falsifiable prediction beyond the internal benchmarks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dual Anchors, Do It Better: Hierarchical Group Merging for Zero-Shot Anomaly Detection." pith.science (2026). https://pith.science/paper/CJAYLOI3

@misc{pith2026260811933,
  author       = {Pith},
  title        = {Pith review of: Dual Anchors, Do It Better: Hierarchical Group Merging for Zero-Shot Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CJAYLOI3}},
  note         = {Machine review of arXiv:2608.11933}
}
read the original abstract

Zero-shot anomaly detection (ZSAD) aims to identify anomalies in unseen domains, a setting that is particularly critical for industrial and medical applications where domain shifts are prevalent. However, most CLIP-based ZSAD methods anchor semantics solely on the text modality, making performance highly sensitive to prompt design and leading to weak visual grounding. To mitigate these limitations, we propose a Dual-Anchor framework that complements conventional text anchors with hierarchical image anchors constructed via a top-down grouping mechanism. This mechanism progressively aggregates local-to-global image features to form normal and abnormal group tokens, which serve as image anchors and act as gating signals in a Group-Gated Token Refiner to enhance the global representation. The refined image anchors are then fused with text prompts to construct dynamic state prompts. By jointly reinforcing visual and textual semantics, our framework stabilizes image-text alignment, reduces prompt dependency, and achieves strong generalization across 8 industrial and 6 medical benchmarks.

Figures

Figures reproduced from arXiv: 2608.11933 by the authors.

Figure 1
Figure 1. Overview of the proposed Dual-Anchor for Zero-Shot Anomaly Detection. Previous methods perform text-driven alignment, anchoring semantics only on text prompts, which causes excessive text dependency and weak visual grounding. Our approach introduces a Hierarchical Group Merging that constructs image semantic anchors and fuses them with text anchors, forming a dual-anchor paradigm that achieves balanced alignment and… view at source ↗
Figure 2
Figure 2. Overall Architecture of our framework. Our framework adopt frozen DINOv3 backbone (Sec. 3.2) to extract multi-scale patch tokens as visual representations. Subsequently, the Hierarchical Group Merging module aggregates these features with learnable group tokens to construct image anchors (Sec. 3.3). The generated image anchors are then utilized by the Group-Gated Token Refiner and the Dynamic State Prompt (Sec. 3.4)… view at source ↗
Figure 3
Figure 3. T-SNE visualization of image and text anchors on VisA and MVTec-AD datasets. Update. The update stage is applied only in the final Merge Block, where the group tokens encapsulating hier￾archical semantics are assigned back to the patch tokens to produce the final assigned representation Fassign. Image Anchor. Through this top-down hierarchical merging process, information from different levels is pro￾gressively inte… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of anomaly localization results on industrial and medical datasets.Our method achieves more precise and consistent anomaly detection. 6 medical-domain datasets. Specifically, the industrial datasets include MVTec AD [2], VisA [31], MPDD [16], BTA…
Figure 5
Figure 5. Figure 5: further visualizes the assignment attention maps across merge stages in an unseen domain. The group to￾kens initially capture diverse features, but progressively form normal/anomaly specific groups, with the final stage closely aligning with the ground-truth masks. The…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 19 canonical work pages

  1. [1]

    Zero-shot versus many-shot: Unsupervised texture anomaly detection

    Toshimichi Aota, Lloyd Teh Tzer Tong, and Takayuki Okatani. Zero-shot versus many-shot: Unsupervised texture anomaly detection. InProceedings of the IEEE/CVF Win- ter Conference on Applications of Computer Vision, pages 5564–5572, 2023. 7

    work page 2023

  2. [2]

    The mvtec anomaly detection dataset: a comprehensive real-world dataset for unsupervised anomaly detection.International Journal of Computer Vi- sion, 129(4):1038–1059, 2021

    Paul Bergmann, Kilian Batzner, Michael Fauser, David Sat- tlegger, and Carsten Steger. The mvtec anomaly detection dataset: a comprehensive real-world dataset for unsupervised anomaly detection.International Journal of Computer Vi- sion, 129(4):1038–1059, 2021. 7

    work page 2021

  3. [3]

    Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs

    Jorge Bernal, F Javier S ´anchez, Gloria Fern´andez-Esparrach, Debora Gil, Cristina Rodr ´ıguez, and Fernando Vilari ˜no. Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physi- cians.Computerized medical imaging and graphics, 43: 99–111, 2015. 7

    work page 2015

  4. [4]

    To- ken merging: Your vit but faster.arXiv preprint arXiv:2210.09461, 2022

    Daniel Bolya, Cheng-Yang Fu, Xiaoliang Dai, Peizhao Zhang, Christoph Feichtenhofer, and Judy Hoffman. To- ken merging: Your vit but faster.arXiv preprint arXiv:2210.09461, 2022. 3, 4

    arXiv 2022

  5. [5]

    Mixed supervision for surface-defect detection: From weakly to fully supervised learning.Computers in Industry, 129: 103459, 2021

    Jakob Bo ˇziˇc, Domen Tabernik, and Danijel Sko ˇcaj. Mixed supervision for surface-defect detection: From weakly to fully supervised learning.Computers in Industry, 129: 103459, 2021. 7

    work page 2021

  6. [6]

    Adaclip: Adapting clip with hybrid learnable prompts for zero-shot anomaly de- tection

    Yunkang Cao, Jiangning Zhang, Luca Frittoli, Yuqi Cheng, Weiming Shen, and Giacomo Boracchi. Adaclip: Adapting clip with hybrid learnable prompts for zero-shot anomaly de- tection. InEuropean Conference on Computer Vision, pages 55–72. Springer, 2024. 2, 6

    work page 2024

  7. [7]

    Xuhai Chen, Yue Han, and Jiangning Zhang. April-gan: A zero-/few-shot anomaly classification and segmentation method for cvpr 2023 vand workshop challenge tracks 1&2: 1st place on zero-shot ad and 4th place on few-shot ad.arXiv preprint arXiv:2305.17382, 2023. 2, 6

    arXiv 2023

  8. [8]

    Clip-ad: A language-guided staged dual-path model for zero-shot anomaly detection

    Xuhai Chen, Jiangning Zhang, Guanzhong Tian, Haoyang He, Wuhao Zhang, Yabiao Wang, Chengjie Wang, and Yong Liu. Clip-ad: A language-guided staged dual-path model for zero-shot anomaly detection. InInternational Joint Confer- ence on Artificial Intelligence, pages 17–33. Springer, 2024. 2

    work page 2024

Show all 31 references
  1. [9]

    Bootstrap fine-grained vision-language alignment for unified zero-shot anomaly localization.arXiv preprint arXiv:2308.15939, 2023

    Hanqiu Deng, Zhaoxiang Zhang, Jinan Bao, and Xingyu Li. Bootstrap fine-grained vision-language alignment for unified zero-shot anomaly localization.arXiv preprint arXiv:2308.15939, 2023. 2

    2023 arXiv

  2. [10]

    Hgformer: Hierarchical grouping transformer for domain generalized semantic segmentation

    Jian Ding, Nan Xue, Gui-Song Xia, Bernt Schiele, and Dengxin Dai. Hgformer: Hierarchical grouping transformer for domain generalized semantic segmentation. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15413–15423, 2023. 3

    2023

  3. [11]

    Multi- task learning for thyroid nodule segmentation with thyroid region prior

    Haifan Gong, Guanqi Chen, Ranran Wang, Xiang Xie, Mingzhi Mao, Yizhou Yu, Fei Chen, and Guanbin Li. Multi- task learning for thyroid nodule segmentation with thyroid region prior. In2021 IEEE 18th international symposium on biomedical imaging (ISBI), pages 257–261. IEEE, 2021. 7

    2021

  4. [12]

    David Gutman, Noel CF Codella, Emre Celebi, Brian Helba, Michael Marchetti, Nabin Mishra, and Allan Halpern. Skin lesion analysis toward melanoma detection: A challenge at the international symposium on biomedical imaging (isbi) 2016, hosted by the international skin imaging c...

    2016 arXiv

  5. [13]

    The endotect 2020 challenge: evaluation and comparison of clas- sification, segmentation and inference time for endoscopy

    Steven A Hicks, Debesh Jha, Vajira Thambawita, P ˚al Halvorsen, Hugo L Hammer, and Michael A Riegler. The endotect 2020 challenge: evaluation and comparison of clas- sification, segmentation and inference time for endoscopy. InInternational Conference on Pattern Recognition, p...

    2020

  6. [14]

    Categorical reparameterization with gumbel-softmax.arXiv preprint arXiv:1611.01144, 2016

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax.arXiv preprint arXiv:1611.01144, 2016. 4

    2016 arXiv

  7. [15]

    Winclip: Zero- /few-shot anomaly classification and segmentation

    Jongheon Jeong, Yang Zou, Taewan Kim, Dongqing Zhang, Avinash Ravichandran, and Onkar Dabeer. Winclip: Zero- /few-shot anomaly classification and segmentation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19606–19616, 2023. 2, 6

    2023

  8. [16]

    Deep learning-based defect detection of metal parts: evaluating current methods in complex condi- tions

    Stepan Jezek, Martin Jonak, Radim Burget, Pavel Dvorak, and Milos Skotak. Deep learning-based defect detection of metal parts: evaluating current methods in complex condi- tions. In2021 13th International congress on ultra mod- ern telecommunications and control systems and wo...

    2021

  9. [17]

    Kvasir-seg: A segmented polyp dataset

    Debesh Jha, Pia H Smedsrud, Michael A Riegler, P ˚al Halvorsen, Thomas De Lange, Dag Johansen, and H˚avard D Johansen. Kvasir-seg: A segmented polyp dataset. InIn- ternational conference on multimedia modeling, pages 451–

  10. [18]

    Aa-clip: Enhancing zero-shot anomaly detection via anomaly-aware clip

    Wenxin Ma, Xu Zhang, Qingsong Yao, Fenghe Tang, Chenxu Wu, Yingtai Li, Rui Yan, Zihang Jiang, and S Kevin Zhou. Aa-clip: Enhancing zero-shot anomaly detection via anomaly-aware clip. InProceedings of the Computer Vi- sion and Pattern Recognition Conference, pages 4744–4754,

  11. [19]

    The concrete distribution: A continuous relaxation of discrete random variables.arXiv preprint arXiv:1611.00712, 2016

    Chris J Maddison, Andriy Mnih, and Yee Whye Teh. The concrete distribution: A continuous relaxation of discrete random variables.arXiv preprint arXiv:1611.00712, 2016. 4

    2016 arXiv

  12. [20]

    Vt-adl: A vision trans- former network for image anomaly detection and localiza- tion

    Pankaj Mishra, Riccardo Verk, Daniele Fornasier, Claudio Piciarelli, and Gian Luca Foresti. Vt-adl: A vision trans- former network for image anomaly detection and localiza- tion. In2021 IEEE 30th International Symposium on Indus- trial Electronics (ISIE), pages 01–06. IEEE, 2021. 7

    2021

  13. [21]

    Vcp-clip: A visual context prompting model for zero-shot anomaly segmenta- tion

    Zhen Qu, Xian Tao, Mukesh Prasad, Fei Shen, Zhengtao Zhang, Xinyi Gong, and Guiguang Ding. Vcp-clip: A visual context prompting model for zero-shot anomaly segmenta- tion. InEuropean Conference on Computer Vision, pages 301–317. Springer, 2024. 2

    2024

  14. [22]

    Bayesian prompt flow learning for zero-shot anomaly detec- tion

    Zhen Qu, Xian Tao, Xinyi Gong, Shichen Qu, Qiyu Chen, Zhengtao Zhang, Xingang Wang, and Guiguang Ding. Bayesian prompt flow learning for zero-shot anomaly detec- tion. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 30398–30408, 2025. 2, 6, 7

    2025

  15. [23]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, p...

    2021

  16. [24]

    Tokenlearner: What can 8 learned tokens do for images and videos?arXiv preprint arXiv:2106.11297, 2021

    Michael S Ryoo, AJ Piergiovanni, Anurag Arnab, Mostafa Dehghani, and Anelia Angelova. Tokenlearner: What can 8 learned tokens do for images and videos?arXiv preprint arXiv:2106.11297, 2021. 3

    2021 arXiv

  17. [25]

    Dinov3.arXiv preprint arXiv:2508.10104, 2025

    Oriane Sim ´eoni, Huy V V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Micha ¨el Ramamonjisoa, et al. Dinov3.arXiv preprint arXiv:2508.10104, 2025. 4

    2025 arXiv

  18. [26]

    Automated polyp detection in colonoscopy videos using shape and context information.IEEE transactions on medi- cal imaging, 35(2):630–644, 2015

    Nima Tajbakhsh, Suryakanth R Gurudu, and Jianming Liang. Automated polyp detection in colonoscopy videos using shape and context information.IEEE transactions on medi- cal imaging, 35(2):630–644, 2015. 7

    2015

  19. [27]

    Weakly supervised learn- ing for industrial optical inspection

    Matthias Wieler and Tobias Hahn. Weakly supervised learn- ing for industrial optical inspection. InDAGM symposium in, page 11, 2007. 7

    2007

  20. [28]

    Groupvit: Semantic segmentation emerges from text supervision

    Jiarui Xu, Shalini De Mello, Sifei Liu, Wonmin Byeon, Thomas Breuel, Jan Kautz, and Xiaolong Wang. Groupvit: Semantic segmentation emerges from text supervision. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 18134–18144, 2022. 3

    2022

  21. [29]

    A coarse-to-fine model for rail surface defect detection.IEEE Transactions on In- strumentation and Measurement, 68(3):656–666, 2018

    Haomin Yu, Qingyong Li, Yunqiang Tan, Jinrui Gan, Jianzhu Wang, Yangli-ao Geng, and Lei Jia. A coarse-to-fine model for rail surface defect detection.IEEE Transactions on In- strumentation and Measurement, 68(3):656–666, 2018. 7

    2018

  22. [30]

    Anomalyclip: Object-agnostic prompt learn- ing for zero-shot anomaly detection.arXiv preprint arXiv:2310.18961, 2023

    Qihang Zhou, Guansong Pang, Yu Tian, Shibo He, and Jiming Chen. Anomalyclip: Object-agnostic prompt learn- ing for zero-shot anomaly detection.arXiv preprint arXiv:2310.18961, 2023. 2, 6

    2023

  23. [31]

    Spot-the-difference self-supervised pre- training for anomaly detection and segmentation

    Yang Zou, Jongheon Jeong, Latha Pemula, Dongqing Zhang, and Onkar Dabeer. Spot-the-difference self-supervised pre- training for anomaly detection and segmentation. InEu- ropean conference on computer vision, pages 392–408. Springer, 2022. 7

    2022

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.