Pith. sign in

REVIEW 3 major objections 3 minor 48 references

IADGPT: Unified LVLM for Few-Shot Industrial Anomaly Detection, Localization, and Reasoning via In-Context Learning

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A single vision-language model, trained in three progressive stages with in-context learning, is claimed to detect, localize, and explain industrial defects from a few exemplar images.

desk verdict The abstract advertises a vision-language anomaly detection system that simply is not in the body—the full text is a different paper. read the letter →

arxiv 2508.10681 v1 pith:AKMJ5C5H submitted 2025-08-14 cs.CV

Mengyang Zhao , Teng Fu , Haiyang Yu , Ke Niu , Bin Li This is my paper
classification cs.CV
keywords few-shotanomalydetectionindustrialinspectionlargevision-languagemodelin-contextlearninglocalizationreasoningvisualdataset
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

As described in its abstract, IADGPT is a unified large vision-language model designed to handle few-shot industrial anomaly detection, localization, and reasoning in one pass. The authors claim that a three-stage progressive training schedule first teaches basic industrial knowledge, then discrepancy awareness, then few-shot generalization via in-context learning, so the model can inspect novel products from just a few example images. They also claim to derive image- and pixel-level anomaly scores from the model's logits and attention maps, combining them with language output to explain defects. The paper reports considerable gains in detection and competitive results in localization and reasoning, supported by a proposed dataset of 100K images across 400 product categories.

What carries the argument

The three-stage progressive training schedule with in-context learning, together with the logit/attention-map scoring scheme. The first two stages instill industrial knowledge and discrepancy awareness; the third stage trains the model to condition on a few exemplar images. The scoring mechanism converts the LVLM's own outputs into image- and pixel-level anomaly scores alongside the natural-language reasoning.

What would settle it

Look in the published text for the IADGPT model, the three-stage training, and the 100K-image dataset; none appear in the supplied full text, and if the camera-ready release omits them, the claimed performance gains have no empirical grounding.

Watch

Extended reading notes

Core claim

In the author's telling, IADGPT is the first unified framework to bring together detection, localization, and reasoning for few-shot industrial anomaly inspection. Its load-bearing design is a three-stage training strategy: the first stage imparts general industrial knowledge, the second develops the ability to spot discrepancies, and the third trains the model to use a few exemplar images as in-context cues for new products. Anomaly scores are extracted from the model's output logits (image-level) and attention maps (pixel-level), so the language-based reasoning and the numerical scores come from the same forward pass. The authors introduce a large dataset with attribute-level text annotati

Load-bearing premise

The submitted full text is a different cybersecurity paper, so the central claim that IADGPT achieves those gains rests on the unsupported premise that the experiments described in the abstract actually exist and are reported somewhere; in the provided manuscript, no IADGPT model, training, dataset, or evaluation appears.

Editorial extensions

If this is right

  • If the claims hold, a single model could replace separate specialist systems for detection, localization, and explanation in industrial quality control.
  • The in-context training paradigm would let operators adapt to new product lines by supplying a few example images, with no fine-tuning.
  • The 100K-image, 400-category dataset would give the community a large-scale resource for few-shot anomaly research.
  • Deriving anomaly scores from the model's own logits and attention maps would make explanations faithful to the numerical prediction.

Reading between the lines

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

  • The abstract promises a dataset and code release (camera-ready); the utility of the claimed gains will depend on whether those releases let others verify the 100K-image scale and reproduce the benchmarks.
  • A natural extension is to test the same in-context learning recipe on other visual inspection domains, such as medical imaging, where few labeled examples are the norm.
  • The claim that attention maps carry pixel-level anomaly signal is testable even without the full framework: one could compare attention-based localization against ground-truth masks on an existing anomaly benchmark.
  • The three-stage curriculum is a concrete hypothesis about how to teach LVLMs domain competence; an ablation that removes the first two stages would directly test whether progressive training is what enables the few-shot gains.
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 / 3 minor

Summary. The abstract announces IADGPT, a unified large vision-language model for few-shot industrial anomaly detection, localization, and reasoning, supported by a three-stage progressive training strategy, a logits/attention scoring mechanism, and a new 100K-image, 400-category industrial dataset. The full text that follows, however, is a different paper: 'Advancing Autonomous Incident Response: Leveraging LLMs and Cyber Threat Intelligence' (footer arXiv:2508.10677v1 [cs.CR]). The body describes a retrieval-augmented generation (RAG) system for SIEM alert enrichment and incident response, built on VirusTotal queries, a CTI vector database, and GPT-4o, with evaluation on 100 real and 10 simulated alerts using LLM-as-judge plus expert assessment. None of the IADGPT components advertised in the abstract—the model, the three-stage training, the few-shot in-context paradigm, the anomaly scoring mechanism, the industrial dataset, or the anomaly detection/localization/reasoning benchmarks—appear anywhere in the submitted text.

Significance. If the abstract's claim were realized as described, the contribution would be significant: a single LVLM performing few-shot industrial anomaly detection, localization, and reasoning with a large multi-category industrial dataset would advance the field and provide a valuable public resource. However, the submitted manuscript does not contain that contribution. The body is a self-contained RAG/LLM incident-response study with a modest evaluation (110 alerts, LLM-judged metrics). Since the central claim is entirely unsupported by the submitted content, no significance can be assigned to the advertised IADGPT framework. The paper's only assessable contribution is the incident-response system, which is outside the advertised scope and is not described in the abstract.

major comments (3)
  1. [Abstract vs. Full Text] The load-bearing claim of the abstract—that 'we propose a unified framework, IADGPT' and that 'Experiments indicate IADGPT achieves considerable performance gains in anomaly detection'—has no supporting implementation, evaluation, or derivation in the submitted body. Section II describes a RAG-based incident response architecture, and Section III evaluates that system on SIEM alerts. No IADGPT model, no three-stage progressive training, no few-shot in-context paradigm, no logits/attention anomaly scoring, and no industrial anomaly benchmarks are present. The submitted content therefore does not support the abstract, and the central claim is unverifiable from the provided material.
  2. [Section III-Evaluation] Tables II, III, and IV report answer relevance, context relevance, and groundedness scores for SIEM alerts in a cybersecurity incident-response setting. These metrics and the alert dataset have no connection to industrial anomaly detection, localization, or reasoning. There are no AUROC, F1, pixel-level IoU, or similar anomaly-detection metrics, and no comparison to any FS-IAD baseline. Consequently, the claimed 'considerable performance gains' in anomaly detection and 'competitiveness in anomaly localization and reasoning' are not just unproven; they are untested by any experiment in the manuscript.
  3. [Dataset Claim] The abstract promises 'a new dataset comprising 100K images across 400 diverse industrial product categories with extensive attribute-level textual annotations' and states it will be released in camera-ready. The full text contains no description, statistics, samples, annotation protocol, or analysis of such a dataset. The only dataset mentioned is a collection of 100 real and 10 simulated SIEM alerts (Section III-A). The claimed industrial dataset is therefore entirely absent, and no evidence supports its existence or characteristics.
minor comments (3)
  1. [Title/Header Mismatch] The manuscript title and abstract refer to IADGPT and arXiv:2508.10681, but the body's title, authors, and footer correspond to 'Advancing Autonomous Incident Response' (arXiv:2508.10677v1 [cs.CR]). The submission is internally inconsistent at the metadata level.
  2. [Section III-B Evaluation methodology] The evaluation relies on auxiliary LLMs grading LLM-generated responses. The authors note cross-validation by experts for 10 simulated alerts, but the real-alert results (Table II) are only LLM-judged. This is a meaningful limitation for the incident-response study, but it is tangential to the IADGPT claims because the IADGPT experiments are absent.
  3. [Reference and Notation Issues] The body contains typographical issues, such as Equation (1) with a garbled fraction layout, and some informal phrasing (e.g., 'the model can performs standard searches'). These are presentation issues in a manuscript that is, in any case, not the advertised paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the submitted body is a different paper, so the claimed IADGPT derivation is entirely absent; no reduction of predictions to inputs can be identified.

full rationale

The assigned paper (arXiv:2508.10681) is titled 'IADGPT: Unified LVLM for Few-Shot Industrial Anomaly Detection, Localization, and Reasoning via In-Context Learning.' The supplied full text, however, is the manuscript 'Advancing Autonomous Incident Response: Leveraging LLMs and Cyber Threat Intelligence' (arXiv:2508.10677v1 [cs.CR]), a study of retrieval-augmented generation for cybersecurity incident response. The body contains no IADGPT model, no three-stage training, no logits/attention scoring, no 100K-image/400-category dataset, and no industrial-anomaly benchmarks. Consequently, the abstract's central claims about IADGPT's performance have no derivation in the body to audit. Circularity analysis requires exhibiting a specific reduction of a claimed result to its inputs (e.g., a fitted parameter renamed as a prediction, or a uniqueness theorem imported from the authors' own prior work). Here there is no derivation chain at all, so no circular step can be pinpointed. The body's evaluation uses auxiliary LLMs (Mistral-large-2407, Llama-3.1-70B, Llama-3.2-3B) to score responses generated by GPT-4o, with a manual expert check on 10 simulated alerts. This is a weak external anchor and a possible validity concern, but it is not circular in the sense of the seven enumerated patterns: the grading models are distinct from the generation model, and the expert provides an independent albeit small validation. No parameter is fitted to a subset and then used to predict a closely related quantity; no self-citation is load-bearing; no known result is renamed. Therefore, despite the severe mismatch between abstract and body, the paper does not exhibit circularity as defined. The honest finding is no significant circularity (score 0), with the caveat that the claimed IADGPT results are unverifiable from the supplied text for reasons of incompleteness rather than circularity.

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

The central claim cannot be audited because its components are absent from the body. The ledger entries mark the load-bearing premises the abstract states but the full text does not support: the existence and adequacy of the 100K/400-category dataset, the transfer promise of three-stage training plus in-context learning, and the structural premise that the body belongs to the advertised paper. No free parameters can be audited because no method details are present. No new physical or conceptual entities are introduced beyond the named system.

assumptions (3)
  • domain assumption A 100K-image, 400-category dataset with attribute-level text annotations exists and is adequate to teach general industrial knowledge.
    The abstract's three-stage training narrative depends on this dataset, but no part of the full text describes, samples, or analyzes it.
  • ad hoc to paper Progressive three-stage training plus in-context exemplars transfers to novel industrial products.
    This is the paper's own transfer premise; it is stated in the abstract and demonstrated nowhere in the full text.
  • ad hoc to paper The supplied full text is the body of the advertised IADGPT paper.
    The entire evaluation depends on this premise. The provided text is a different paper with arXiv footer 2508.10677v1 [cs.CR], so the premise fails.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IADGPT: Unified LVLM for Few-Shot Industrial Anomaly Detection, Localization, and Reasoning via In-Context Learning." pith.science (2026). https://pith.science/paper/AKMJ5C5H

@misc{pith2026250810681,
  author       = {Pith},
  title        = {Pith review of: IADGPT: Unified LVLM for Few-Shot Industrial Anomaly Detection, Localization, and Reasoning via In-Context Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AKMJ5C5H}},
  note         = {Machine review of arXiv:2508.10681}
}
read the original abstract

Few-Shot Industrial Anomaly Detection (FS-IAD) has important applications in automating industrial quality inspection. Recently, some FS-IAD methods based on Large Vision-Language Models (LVLMs) have been proposed with some achievements through prompt learning or fine-tuning. However, existing LVLMs focus on general tasks but lack basic industrial knowledge and reasoning capabilities related to FS-IAD, making these methods far from specialized human quality inspectors. To address these challenges, we propose a unified framework, IADGPT, designed to perform FS-IAD in a human-like manner, while also handling associated localization and reasoning tasks, even for diverse and novel industrial products. To this end, we introduce a three-stage progressive training strategy inspired by humans. Specifically, the first two stages gradually guide IADGPT in acquiring fundamental industrial knowledge and discrepancy awareness. In the third stage, we design an in-context learning-based training paradigm, enabling IADGPT to leverage a few-shot image as the exemplars for improved generalization to novel products. In addition, we design a strategy that enables IADGPT to output image-level and pixel-level anomaly scores using the logits output and the attention map, respectively, in conjunction with the language output to accomplish anomaly reasoning. To support our training, we present a new dataset comprising 100K images across 400 diverse industrial product categories with extensive attribute-level textual annotations. Experiments indicate IADGPT achieves considerable performance gains in anomaly detection and demonstrates competitiveness in anomaly localization and reasoning. We will release our dataset in camera-ready.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 30 canonical work pages

  1. [1]

    Bai, J.; Bai, S.; Yang, S.; Wang, S.; Tan, S.; Wang, P.; Lin, J.; Zhou, C.; and Zhou, J. 2023. Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond. arXiv:2308.12966

    arXiv 2023

  2. [2]

    Bai, S.; Chen, K.; Liu, X.; Wang, J.; Ge, W.; Song, S.; Dang, K.; Wang, P.; Wang, S.; Tang, J.; Zhong, H.; Zhu, Y.; Yang, M.; Li, Z.; Wan, J.; Wang, P.; Ding, W.; Fu, Z.; Xu, Y.; Ye, J.; Zhang, X.; Xie, T.; Cheng, Z.; Zhang, H.; Yang, Z.; Xu, H.; and Lin, J. 2025. Qwen2.5-VL Technical Report. arXiv preprint arXiv:2502.13923

    arXiv 2025

  3. [3]

    Bergmann, P.; Batzner, K.; Fauser, M.; Sattlegger, D.; and Steger, C. 2021. The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection. Int. J. Comput. Vis., 129(4): 1038--1059

    work page 2021

  4. [4]

    Bonfiglioli, L.; Toschi, M.; Silvestri, D.; Fioraio, N.; and Gregorio, D. D. 2022. The Eyecandies Dataset for Unsupervised Multimodal Anomaly Detection and Localization. In Wang, L.; Gall, J.; Chin, T.; Sato, I.; and Chellappa, R., eds., Computer Vision - ACCV 2022 - 16th Asian Conference on Computer Vision, Macao, China, December 4-8, 2022, Proceedings, ...

    work page 2022

  5. [5]

    Bozic, J.; Tabernik, D.; and Skocaj, D. 2021. Mixed supervision for surface-defect detection: From weakly to fully supervised learning. Comput. Ind., 129: 103459

    work page 2021

  6. [6]

    Cao, Y.; Xu, X.; Sun, C.; Huang, X.; and Shen, W. 2023. Towards Generic Anomaly Detection and Understanding: Large-scale Visual-linguistic Model (GPT-4V) Takes the Lead. CoRR, abs/2311.02782

    arXiv 2023

  7. [7]

    Chen, X.; Han, Y.; and Zhang, J. 2023. 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

    arXiv 2023

  8. [8]

    Dai, W.; Li, J.; Li, D.; Tiong, A. M. H.; Zhao, J.; Wang, W.; Li, B.; Fung, P.; and Hoi, S. C. H. 2023. InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning. In Oh, A.; Naumann, T.; Globerson, A.; Saenko, K.; Hardt, M.; and Levine, S., eds., Advances in Neural Information Processing Systems 36: Annual Conference on Neural I...

    work page 2023

Show all 48 references
  1. [9]

    Defard, T.; Setkov, A.; Loesch, A.; and Audigier, R. 2020. PaDiM: A Patch Distribution Modeling Framework for Anomaly Detection and Localization. In Bimbo, A. D.; Cucchiara, R.; Sclaroff, S.; Farinella, G. M.; Mei, T.; Bertini, M.; Escalante, H. J.; and Vezzani, R., eds., Patt...

    2020

  2. [10]

    Fang, Z.; Wang, X.; Li, H.; Liu, J.; Hu, Q.; and Xiao, J. 2023. FastRecon: Few-shot Industrial Anomaly Detection via Fast Feature Reconstruction. In IEEE/CVF International Conference on Computer Vision, ICCV 2023, Paris, France, October 1-6, 2023 , 17435--17444. IEEE

    2023

  3. [11]

    Gao, B. 2024. MetaUAS: Universal Anomaly Segmentation with One-Prompt Meta-Learning. In Globersons, A.; Mackey, L.; Belgrave, D.; Fan, A.; Paquet, U.; Tomczak, J. M.; and Zhang, C., eds., Advances in Neural Information Processing Systems 38: Annual Conference on Neural Informa...

    2024

  4. [12]

    Gu, Z.; Zhu, B.; Zhu, G.; Chen, Y.; Tang, M.; and Wang, J. 2024. AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models. In Wooldridge, M. J.; Dy, J. G.; and Natarajan, S., eds., Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty...

    2024

  5. [13]

    Jeong, J.; Zou, Y.; Kim, T.; Zhang, D.; Ravichandran, A.; and Dabeer, O. 2023. WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023, Vancouver, BC, Canada, June 17-24, 2023 , 19606--19616. IEEE

    2023

  6. [14]

    C.; Lo, W.-Y.; Doll \'a r, P.; and Girshick, R

    Kirillov, A.; Mintun, E.; Ravi, N.; Mao, H.; Rolland, C.; Gustafson, L.; Xiao, T.; Whitehead, S.; Berg, A. C.; Lo, W.-Y.; Doll \'a r, P.; and Girshick, R. 2023. Segment Anything. arXiv:2304.02643

    2023 arXiv

  7. [15]

    Li, C.-L.; Sohn, K.; Yoon, J.; and Pfister, T. 2021 a . CutPaste: Self-Supervised Learning for Anomaly Detection and Localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 9664--9674

    2021

  8. [16]

    Li, F.; Zhang, R.; Zhang, H.; Zhang, Y.; Li, B.; Li, W.; Ma, Z.; and Li, C. 2024 a . LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models. CoRR, abs/2407.07895

    2024 arXiv

  9. [17]

    Li, N.; Jiang, K.; Ma, Z.; Wei, X.; Hong, X.; and Gong, Y. 2021 b . Anomaly Detection Via Self-Organizing Map. In 2021 IEEE International Conference on Image Processing, ICIP 2021, Anchorage, AK, USA, September 19-22, 2021 , 974--978. IEEE

    2021

  10. [18]

    Li, X.; Sun, X.; Meng, Y.; Liang, J.; Wu, F.; and Li, J. 2019. Dice loss for data-imbalanced NLP tasks. arXiv preprint arXiv:1911.02855

    2019 arXiv

  11. [19]

    Li, X.; Zhang, Z.; Tan, X.; Chen, C.; Qu, Y.; Xie, Y.; and Ma, L. 2024 b . PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, 2024 , 168...

    2024

  12. [20]

    Li, Y.; Zhang, S.; Li, K.; and Lao, Q. 2024 c . One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection. In Globersons, A.; Mackey, L.; Belgrave, D.; Fan, A.; Paquet, U.; Tomczak, J. M.; and Zhang, C., eds., Advances in Neural Information Processing Systems 38: A...

    2024

  13. [21]

    Lin, B.; Ye, Y.; Zhu, B.; Cui, J.; Ning, M.; Jin, P.; and Yuan, L. 2023. Video-LLaVA: Learning United Visual Representation by Alignment Before Projection. CoRR, abs/2311.10122

    2023 arXiv

  14. [22]

    B.; He, K.; and Doll \' a r, P

    Lin, T.; Goyal, P.; Girshick, R. B.; He, K.; and Doll \' a r, P. 2017. Focal Loss for Dense Object Detection. In IEEE International Conference on Computer Vision, ICCV 2017, Venice, Italy, October 22-29, 2017 , 2999--3007. IEEE Computer Society

    2017

  15. [23]

    Liu, H.; Li, C.; Wu, Q.; and Lee, Y. J. 2023. Visual Instruction Tuning

    2023

  16. [24]

    Loshchilov, I.; and Hutter, F. 2017. SGDR: Stochastic Gradient Descent with Warm Restarts. In 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings . OpenReview.net

    2017

  17. [25]

    Loshchilov, I.; and Hutter, F. 2019. Decoupled Weight Decay Regularization. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019 . OpenReview.net

    2019

  18. [26]

    Lv, W.; Su, Q.; and Xu, W. 2025. One-for-All Few-Shot Anomaly Detection via Instance-Induced Prompt Learning. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025 . OpenReview.net

    2025

  19. [27]

    Mishra, P.; Verk, R.; Fornasier, D.; Piciarelli, C.; and Foresti, G. L. 2021. VT-ADL: A Vision Transformer Network for Image Anomaly Detection and Localization. In 30th IEEE International Symposium on Industrial Electronics, ISIE 2021, Kyoto, Japan, June 20-23, 2021 , 1--6. IEEE

    2021

  20. [28]

    OpenAI; Achiam, J.; Adler, S.; Agarwal, S.; and Lama Ahmad, e. a. 2024. GPT-4 Technical Report. arXiv:2303.08774

    2024 arXiv

  21. [29]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, 8748--8763. PMLR

    2021

  22. [30]

    Reimers, N.; and Gurevych, I. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Inui, K.; Jiang, J.; Ng, V.; and Wan, X., eds., Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Confere...

    2019

  23. [31]

    Roth, K.; Pemula, L.; Zepeda, J.; Sch \" o lkopf, B.; Brox, T.; and Gehler, P. V. 2022. Towards Total Recall in Industrial Anomaly Detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022 , 14298--14308. IEEE

    2022

  24. [32]

    Tabernik, D.; Sela, S.; Skvarc, J.; and Skocaj, D. 2020. Segmentation-based deep-learning approach for surface-defect detection. J. Intell. Manuf., 31(3): 759--776

    2020

  25. [33]

    Tamura, M. 2023. Random Word Data Augmentation with CLIP for Zero-Shot Anomaly Detection. In 34th British Machine Vision Conference 2023, BMVC 2023, Aberdeen, UK, November 20-24, 2023 , 18--21. BMVA Press

    2023

  26. [34]

    Tao, F.; Xie, G.; Zhao, F.; and Shu, X. 2025. Kernel-Aware Graph Prompt Learning for Few-Shot Anomaly Detection. In Walsh, T.; Shah, J.; and Kolter, Z., eds., AAAI-25, Sponsored by the Association for the Advancement of Artificial Intelligence, February 25 - March 4, 2025, Phi...

    2025

  27. [35]

    Wang, C.; Zhu, W.; Gao, B.; Gan, Z.; Zhang, J.; Gu, Z.; Qian, S.; Chen, M.; and Ma, L. 2024 a . Real-IAD: A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detection. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seat...

    2024

  28. [36]

    Wang, P.; Bai, S.; Tan, S.; Wang, S.; Fan, Z.; Bai, J.; Chen, K.; Liu, X.; Wang, J.; Ge, W.; Fan, Y.; Dang, K.; Du, M.; Ren, X.; Men, R.; Liu, D.; Zhou, C.; Zhou, J.; and Lin, J. 2024 b . Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution. Co...

    2024 arXiv

  29. [37]

    Wei, H.; Liu, C.; Chen, J.; Wang, J.; Kong, L.; Xu, Y.; Ge, Z.; Zhao, L.; Sun, J.; Peng, Y.; et al. 2024. General OCR Theory: Towards OCR-2.0 via a Unified End-to-end Model. arXiv preprint arXiv:2409.01704

    2024 arXiv

  30. [38]

    Wieler, M.; and Hahn, T. 2007. Weakly supervised learning for industrial optical inspection. In DAGM symposium in, volume 6, 11

    2007

  31. [39]

    M.; and Dwivedi, I

    Xu, J.; Lo, S.-Y.; Safaei, B.; Patel, V. M.; and Dwivedi, I. 2025. Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models. arXiv preprint arXiv:2502.07601

    2025 arXiv

  32. [40]

    Yun, E.; Ha, H.; Nam, Y.; and Lee, B. D. 2025. Language-Assisted Feature Transformation for Anomaly Detection. In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025 . OpenReview.net

    2025

  33. [41]

    Zhang, J.; He, H.; Chen, X.; Xue, Z.; Wang, Y.; Wang, C.; Xie, L.; and Liu, Y. 2024 a . GPT-4V-AD: Exploring Grounding Potential of VQA-oriented GPT-4V for Zero-shot Anomaly Detection. arXiv:2311.02612

    2024 arXiv

  34. [42]

    Zhang, T.; Zhong, S.; Xu, W.; Yan, L.; and Zou, X. 2024 b . Catenary Insulator Defect Detection: A Dataset and an Unsupervised Baseline. IEEE Transactions on Instrumentation and Measurement, 73: 1--15

    2024

  35. [43]

    Zhao, H.; Cai, Z.; Si, S.; Ma, X.; An, K.; Chen, L.; Liu, Z.; Wang, S.; Han, W.; and Chang, B. 2024. MMICL: Empowering Vision-language Model with Multi-Modal In-Context Learning. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, M...

    2024

  36. [44]

    Zhou, Q.; Pang, G.; Tian, Y.; He, S.; and Chen, J. 2024. AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024 . OpenReview.net

    2024

  37. [45]

    Zhu, D.; Chen, J.; Shen, X.; Li, X.; and Elhoseiny, M. 2023. MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models. arXiv preprint arXiv:2304.10592

    2023 arXiv

  38. [46]

    Zhu, J.; and Pang, G. 2024. Toward Generalist Anomaly Detection via In-Context Residual Learning with Few-Shot Sample Prompts. In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 17826--17836

    2024

  39. [47]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  40. [48]

    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 gl...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.