arxiv: 2604.07097 · v1 · submitted 2026-04-08 · 💻 cs.CV

Recognition: no theorem link

Novel Anomaly Detection Scenarios and Evaluation Metrics to Address the Ambiguity in the Definition of Normal Samples

Reiji Saito , Satoshi Kamiya , Kazuhiro Hotta

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:24 UTC · model grok-4.3

classification 💻 cs.CV

keywords anomaly detectionnormal sample ambiguityRePasteMVTec ADevaluation metricsindustrial anomaly detectionspecification changesimage re-pasting

0 comments

The pith

RePaste improves handling of ambiguous normal samples in anomaly detection through iterative region re-pasting on new scenarios and metrics.

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

Standard anomaly detection assumes clear normal samples for training, yet in practice the boundary is fuzzy, as when minor defects are tolerable or not depending on current specifications. The paper creates scenarios on the MVTec AD benchmark that vary which small defects count as normal to mimic real industrial shifts in requirements. It also defines a new evaluation metric sensitive to this ambiguity. RePaste addresses the issue by taking areas that scored high as anomalous in one pass and pasting them into the training input for the next pass, forcing the model to learn finer distinctions. This combination yields top results on the new metric without losing ground on conventional AUROC and PRO measures.

Core claim

The paper establishes that novel scenarios simulating specification changes in normal definitions, paired with a dedicated evaluation metric, allow the RePaste approach—iteratively re-pasting high-anomaly-score regions from prior steps into inputs for subsequent steps—to achieve state-of-the-art performance on the proposed metric using the MVTec AD benchmark, all while preserving high AUROC and PRO scores.

What carries the argument

RePaste, the iterative re-pasting of regions with high anomaly scores from the previous step into the input for the next step, which refines the model's handling of borderline normal samples.

If this is right

Models trained this way perform better when the definition of acceptable defects changes over time in manufacturing.
Evaluation can now account for varying precision needs instead of assuming a fixed normal class.
RePaste keeps strong scores on traditional metrics such as AUROC and PRO while improving on the ambiguity metric.
The method works on the MVTec AD dataset across the introduced scenarios.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Applying RePaste could cut down on retraining cycles when production standards are updated.
Similar iterative pasting might help in other semi-supervised tasks with fuzzy boundaries.
Real factory deployments with operator feedback on ambiguous cases would provide stronger validation than benchmark-only tests.
The approach highlights that focusing training on uncertain regions can improve robustness to definition shifts.

Load-bearing premise

That the invented scenarios and metric truly mirror the ambiguity found in actual industrial settings, and that re-pasting high-anomaly regions improves performance without creating artifacts or skewing the learning process.

What would settle it

Running RePaste on a dataset of real industrial images where experts have explicitly labeled which minor defects should count as normal or anomalous under different specs, and finding no improvement over standard methods on the proposed metric.

Figures

Figures reproduced from arXiv: 2604.07097 by Kazuhiro Hotta, Reiji Saito, Satoshi Kamiya.

**Figure 2.** Figure 2: A2N includes “Anomaly-to-Normal sub-scenario” [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: N2A includes “Normal-to-Anomaly sub-scenario” [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Examples where the anomaly score map shows a high [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Overview of Repaste. To detect small scratches and [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: Qualitative visualization for pixel-level anomaly seg [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 7.** Figure 7: The difference between RePaste with and without [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

read the original abstract

In conventional anomaly detection, training data consist of only normal samples. However, in real-world scenarios, the definition of a normal sample is often ambiguous. For example, there are cases where a sample has small scratches or stains but is still acceptable for practical usage. On the other hand, higher precision is required when manufacturing equipment is upgraded. In such cases, normal samples may include small scratches, tiny dust particles, or a foreign object that we would prefer to classify as an anomaly. Such cases frequently occur in industrial settings, yet they have not been discussed until now. Thus, we propose novel scenarios and an evaluation metric to accommodate specification changes in real-world applications. Furthermore, to address the ambiguity of normal samples, we propose the RePaste, which enhances learning by re-pasting regions with high anomaly scores from the previous step into the input for the next step. On our scenarios using the MVTec AD benchmark, RePaste achieved the state-of-the-art performance with respect to the proposed evaluation metric, while maintaining high AUROC and PRO scores. Code: https://github.com/ReijiSoftmaxSaito/Scenario

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper flags a real industrial gap around fuzzy normal definitions, proposes scenarios plus a custom metric on MVTec, and offers RePaste as an iterative fix that holds up on standard scores.

read the letter

The core point is that conventional anomaly detection assumes a crisp normal class, but factories often have shifting tolerances for minor defects like small scratches. The authors build scenarios on MVTec AD that vary which defects count as normal, introduce a metric to score performance under those conditions, and describe RePaste: an iterative step that takes high-anomaly regions from one pass and re-pastes them into the next input to sharpen the boundary around ambiguous cases. Their results show the method leading on the new metric while preserving high AUROC and PRO, and they release code, which helps with checking the claims. That motivation lands because it matches actual deployment headaches when specs tighten or loosen. The scenarios and metric are the clearest addition; RePaste itself is a straightforward loop but applied in a way that seems tailored to the ambiguity problem. Soft spots are mainly around transparency. The abstract leaves the exact scenario construction and metric formula light on detail, so it is not obvious whether the metric is robust or inadvertently tuned to reward re-pasting. RePaste could also accumulate early errors if the first anomaly map is noisy, and without ablations against simpler baselines or other uncertainty methods it is hard to judge how much the iteration really adds versus just extra training. The SOTA claim is scoped to their metric, which is honest but narrows the takeaway. This is useful reading for people building anomaly detectors for manufacturing lines where quality thresholds are not fixed. A practitioner who needs to adapt models to changing specs would get concrete ideas from the scenarios. It is worth sending to peer review because the practical framing is sound, the code is public, and the experiments can be examined directly even if the metric needs more justification in revision.

Referee Report

2 major / 2 minor

Summary. The paper argues that standard anomaly detection assumes unambiguous normal samples, but real industrial applications often involve ambiguous cases (e.g., small scratches or dust that may be acceptable or unacceptable depending on evolving specifications). It introduces novel scenarios and a custom evaluation metric on the MVTec AD benchmark to capture such ambiguity, along with the RePaste method that iteratively re-pastes high-anomaly-score regions from prior steps into the next training input. Experiments claim that RePaste reaches state-of-the-art on the new metric while retaining high AUROC and PRO scores.

Significance. If the scenarios and metric are shown to meaningfully represent specification-driven ambiguity and RePaste's iterative mechanism demonstrably improves handling without artifacts, the work could offer a practical framework for adapting anomaly detectors to changing industrial requirements. The preservation of standard metrics alongside the new one and the public code release are strengths that support reproducibility and broader adoption.

major comments (2)

[metric definition section] The section defining the proposed evaluation metric does not provide a formal justification or derivation showing why it better captures ambiguity in normal samples compared to AUROC/PRO; without this, the SOTA claim on the new metric is difficult to interpret as an advance rather than a reweighting of existing signals.
[experimental scenarios] In the experimental results on MVTec AD, the construction of the novel scenarios (how specific defects are selected or re-labeled to simulate ambiguity) lacks sufficient detail on selection criteria and coverage across categories, which is load-bearing for validating that RePaste addresses the stated real-world problem rather than an artificial subset.

minor comments (2)

[method section] The description of the RePaste iterative procedure would benefit from pseudocode or a clear diagram illustrating the re-pasting step and how anomaly scores are thresholded across iterations.
[figures] Figure captions and axis labels in the results section should explicitly state which scenarios and metric are being plotted to avoid ambiguity for readers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's constructive feedback on our manuscript. We address each major comment below and will revise the paper to incorporate clarifications and additional details as outlined.

read point-by-point responses

Referee: [metric definition section] The section defining the proposed evaluation metric does not provide a formal justification or derivation showing why it better captures ambiguity in normal samples compared to AUROC/PRO; without this, the SOTA claim on the new metric is difficult to interpret as an advance rather than a reweighting of existing signals.

Authors: We thank the referee for this observation. The metric is specifically formulated to quantify detection performance when normal samples include varying degrees of acceptable defects (e.g., by weighting false positives on small ambiguous anomalies differently from clear anomalies), which AUROC and PRO treat uniformly. This design directly addresses specification-driven changes rather than reweighting existing signals. To strengthen the presentation, we will add a short derivation subsection explaining the metric's construction from the ambiguity scenarios and its divergence from standard metrics. We will also include a brief comparison table showing how the new metric ranks methods differently under ambiguity. revision: yes
Referee: [experimental scenarios] In the experimental results on MVTec AD, the construction of the novel scenarios (how specific defects are selected or re-labeled to simulate ambiguity) lacks sufficient detail on selection criteria and coverage across categories, which is load-bearing for validating that RePaste addresses the stated real-world problem rather than an artificial subset.

Authors: We agree that expanded details are required to demonstrate the scenarios' grounding in real industrial ambiguity. In the revised manuscript, we will add a new subsection in the experiments that specifies: (i) selection criteria (defect size thresholds and types drawn from MVTec annotations, with small scratches/stains re-labeled as normal to simulate loose specifications), (ii) re-labeling procedure per scenario, and (iii) coverage statistics (all 15 categories represented, with per-category counts and ambiguity levels). These additions will confirm the scenarios are not artificial but representative of the problem stated in the introduction. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper introduces novel scenarios for ambiguous normal samples on MVTec AD, a custom evaluation metric, and the RePaste iterative procedure that re-pastes high-anomaly regions. No load-bearing step reduces by construction to its own inputs, fitted parameters renamed as predictions, or self-citation chains. The SOTA claim is scoped to the proposed metric and does not rely on external unverified uniqueness theorems or ansatzes smuggled via prior work. The derivation is self-contained against the new benchmark.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides insufficient detail to identify specific free parameters or invented entities; RePaste is presented as a method rather than a new entity.

axioms (1)

domain assumption Iteratively re-pasting high-anomaly regions improves model performance on ambiguous normal samples.
Core assumption underlying the RePaste proposal.

pith-pipeline@v0.9.0 · 5504 in / 1082 out tokens · 41551 ms · 2026-05-10T18:24:38.839230+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

42 extracted references · 4 canonical work pages · 1 internal anchor

[1]

Efficien- tad: Accurate visual anomaly detection at millisecond-level latencies

Kilian Batzner, Lars Heckler, and Rebecca K ¨onig. Efficien- tad: Accurate visual anomaly detection at millisecond-level latencies. InProceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision, pages 128–138,
[2]

Mvtec ad – a comprehensive real-world dataset for unsupervised anomaly detection

Paul Bergmann, Michael Fauser, David Sattlegger, and Carsten Steger. Mvtec ad – a comprehensive real-world dataset for unsupervised anomaly detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 1711–1720, 2019. 6

2019
[3]

Andrew P. Bradley. The use of the area under the roc curve in the evaluation of machine learning algorithms.Pattern Recognition, 30(7):1145–1159, 1997. 2

1997
[4]

Emerg- ing properties in self-supervised vision transformers

Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9650–9660, 2021. 3

2021
[5]

Riemannian walk for incremen- tal learning: Understanding forgetting and intransigence

Arslan Chaudhry, Puneet K Dokania, Thalaiyasingam Ajan- than, and Philip HS Torr. Riemannian walk for incremen- tal learning: Understanding forgetting and intransigence. In Proceedings of the European conference on computer vision (ECCV), pages 532–547, 2018. 2

2018
[6]

A unified anomaly synthesis strategy with gradient ascent for industrial anomaly detection and localization

Qiyu Chen, Huiyuan Luo, Chengkan Lv, and Zhengtao Zhang. A unified anomaly synthesis strategy with gradient ascent for industrial anomaly detection and localization. In Proceedings of the European Conference on Computer Vi- sion, pages 37–54, 2024. 1, 2, 3, 6, 8

2024
[7]

Cimpoi, S

M. Cimpoi, S. Maji, I. Kokkinos, S. Mohamed, , and A. Vedaldi. Describing textures in the wild. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3606–3613, 2014. 3

2014
[8]

Padim: A patch distribution modeling framework for anomaly detection and localization

Thomas Defard, Aleksandr Setkov, Angelique Loesch, and Romaric Audigier. Padim: A patch distribution modeling framework for anomaly detection and localization. InPro- ceedings of the International Conference on Pattern Recog- nition, pages 475–489, 2021. 3

2021
[9]

Anomaly detection via reverse distillation from one-class embedding

Hanqiu Deng and Xingyu Li. Anomaly detection via reverse distillation from one-class embedding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9737–9746, 2022. 2, 3, 8

2022
[10]

A brief review of domain adaptation

Abolfazl Farahani, Sahar V oghoei, Khaled Rasheed, and Hamid R Arabnia. A brief review of domain adaptation. Advances in data science and information engineering: pro- ceedings from ICDATA 2020 and IKE 2020, pages 877–894,

2020
[11]

Mamba: Linear-time sequence mod- eling with selective state spaces

Albert Gu and Tri Dao. Mamba: Linear-time sequence mod- eling with selective state spaces. InProceedings of the First Conference on Language Modeling, 2024. 3, 8

2024
[12]

Cflow-ad: Real-time unsupervised anomaly detection with localization via conditional normalizing flows

Denis Gudovskiy, Shun Ishizaka, and Kazuki Kozuka. Cflow-ad: Real-time unsupervised anomaly detection with localization via conditional normalizing flows. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 98–107, 2022. 3

2022
[13]

Dinomaly: The less is more philosophy in multi-class unsupervised anomaly detection

Jia Guo, Shuai Lu, Weihang Zhang, Fang Chen, Huiqi Li, and Hongen Liao. Dinomaly: The less is more philosophy in multi-class unsupervised anomaly detection. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20405–20415, 2025. 2, 3, 8

2025
[14]

Haoyang He, Yuhu Bai, Jiangning Zhang, Qingdong He, Hongxu Chen, Zhenye Gan, Chengjie Wang, Xiangtai Li, Guanzhong Tian, and Lei Xie. Mambaad: Exploring state space models for multi-class unsupervised anomaly detec- tion.Proceedings of the International Conference on Neural Information Processing Systems, 37:71162–71187, 2024. 3

2024
[15]

A diffusion-based framework for multi-class anomaly detection

Haoyang He, Jiangning Zhang, Hongxu Chen, Xuhai Chen, Zhishan Li, Xu Chen, Yabiao Wang, Chengjie Wang, and Lei Xie. A diffusion-based framework for multi-class anomaly detection. InProceedings of the Association for the Advance- ment of Artificial Intelligence Conference on Artificial Intel- ligence, pages 8472–8480, 2024. 2, 3, 8

2024
[16]

Recon- patch: Contrastive patch representation learning for indus- trial anomaly detection

Jeeho Hyun, Sangyun Kim, Giyoung Jeon, Seung Hwan Kim, Kyunghoon Bae, and Byung Jun Kang. Recon- patch: Contrastive patch representation learning for indus- trial anomaly detection. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2052–2061, 2024. 1

2052
[17]

Universal domain adaptation for robust handling of distributional shifts in nlp

Hyuhng Joon Kim, Hyunsoo Cho, Sang-Woo Lee, Junyeob Kim, Choonghyun Park, Sang-goo Lee, Kang Yoo, and Taeuk Kim. Universal domain adaptation for robust handling of distributional shifts in nlp. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 5888– 5905, 2023. 2

2023
[18]

Anovit: Unsupervised anomaly detection and localization with vision transformer- based encoder-decoder.IEEE Access, 10:46717–46724,

Yunseung Lee and Pilsung Kang. Anovit: Unsupervised anomaly detection and localization with vision transformer- based encoder-decoder.IEEE Access, 10:46717–46724,
[19]

Cutpaste: Self-supervised learning for anomaly de- tection and localization

Chun-Liang Li, Kihyuk Sohn, Jinsung Yoon, and Tomas Pfister. Cutpaste: Self-supervised learning for anomaly de- tection and localization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9664–9674, 2021. 3

2021
[20]

Learning without forgetting

Zhizhong Li and Derek Hoiem. Learning without forgetting. IEEE transactions on pattern analysis and machine intelli- gence, 40(12):2935–2947, 2017. 1

2017
[21]

Beyond not-forgetting: Continual learning with backward knowledge transfer.Advances in Neural Information Processing Sys- tems, 35:16165–16177, 2022

Sen Lin, Li Yang, Deliang Fan, and Junshan Zhang. Beyond not-forgetting: Continual learning with backward knowledge transfer.Advances in Neural Information Processing Sys- tems, 35:16165–16177, 2022. 2

2022
[22]

Deep unsupervised domain adaptation: A review of recent ad- vances and perspectives.arXiv preprint arXiv:2208.07422,

Xiaofeng Liu, Chaehwa Yoo, Fangxu Xing, Hyejin Oh, Georges El Fakhri, Je-Won Kang, and Jonghye Woo. Deep unsupervised domain adaptation: A review of recent ad- vances and perspectives.arXiv preprint arXiv:2208.07422,

work page arXiv
[23]

Simplenet: A simple network for image anomaly detection and localization

Zhikang Liu, Yiming Zhou, Yuansheng Xu, and Zilei Wang. Simplenet: A simple network for image anomaly detection and localization. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pages 20402–20411, 2023. 2, 3, 6, 8

2023
[24]

Learning under concept drift: A review

Jie Lu, Anjin Liu, Fan Dong, Feng Gu, Joao Gama, and Guangquan Zhang. Learning under concept drift: A review. IEEE transactions on knowledge and data engineering, 31 (12):2346–2363, 2018. 1 9

2018
[25]

Exploring intrinsic normal prototypes within a single im- age for universal anomaly detection

Wei Luo, Yunkang Cao, Haiming Yao, Xiaotian Zhang, Jianan Lou, Yuqi Cheng, Weiming Shen, and Wenyong Yu. Exploring intrinsic normal prototypes within a single im- age for universal anomaly detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9974–9983, 2025. 2, 3, 8

2025
[26]

Anomaly detection with conditioned denoising diffusion models

Arian Mousakhan, Thomas Brox, and Jawad Tayyub. Anomaly detection with conditioned denoising diffusion models. InDAGM German Conference on Pattern Recog- nition, pages 181–195, 2024. 3

2024
[27]

Maxime Oquab, Timoth ´ee Darcet, Th´eo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herve Je- gou, Julien Mairal, Patr...

2024
[28]

Variational inference with normalizing flows

Danilo Rezende and Shakir Mohamed. Variational inference with normalizing flows. InProceedings of the International Conference on Machine Learning, pages 1530–1538, 2015. 3

2015
[29]

Towards to- tal recall in industrial anomaly detection

Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Sch¨olkopf, Thomas Brox, and Peter Gehler. Towards to- tal recall in industrial anomaly detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14318–14328, 2022. 2, 3, 6, 8

2022
[30]

J. C. Schlimmer and R. H. Granger Jr. Incremental learning from noisy data.Machine Learning, 1(3):317–354, 1986. 1

1986
[31]

Unseen visual anomaly generation

Han Sun, Yunkang Cao, Hao Dong, and Olga Fink. Unseen visual anomaly generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25508–25517, 2025. 4, 6

2025
[32]

Duong, Chanh D

Tran Dinh Tien, Anh Tuan Nguyen, Nguyen Hoang Tran, Ta Duc Huy, Soan T.M. Duong, Chanh D. Tr. Nguyen, and Steven Q. H. Truong. Revisiting reverse distillation for anomaly detection. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 24511–24520, 2023. 2, 3, 6, 8

2023
[33]

A comprehensive survey of continual learning: Theory, method and application.IEEE transactions on pattern analysis and machine intelligence, 46(8):5362–5383, 2024

Liyuan Wang, Xingxing Zhang, Hang Su, and Jun Zhu. A comprehensive survey of continual learning: Theory, method and application.IEEE transactions on pattern analysis and machine intelligence, 46(8):5362–5383, 2024. 1

2024
[34]

Uninet: A contrastive learning-guided unified framework with feature selection for anomaly detection

Shun Wei, Jielin Jiang, and Xiaolong Xu. Uninet: A contrastive learning-guided unified framework with feature selection for anomaly detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9994–10003, 2025. 2, 3, 8

2025
[35]

Reconstruction student with attention for student-teacher pyramid matching.arXiv preprint arXiv:2111.15376, 2021

Shinji Yamada and Kazuhiro Hotta. Reconstruction student with attention for student-teacher pyramid matching.arXiv preprint arXiv:2111.15376, 2021. 3

work page arXiv 2021
[36]

Memseg: A semi- supervised method for image surface defect detection using differences and commonalities.Engineering Applications of Artificial Intelligence, 119:105835, 2023

Minghui Yang, Peng Wu, and Hui Feng. Memseg: A semi- supervised method for image surface defect detection using differences and commonalities.Engineering Applications of Artificial Intelligence, 119:105835, 2023. 3, 4, 6

2023
[37]

Glad: Towards better recon- struction with global and local adaptive diffusion models for unsupervised anomaly detection

Hang Yao, Ming Liu, Zhicun Yin, Zifei Yan, Xiaopeng Hong, and Wangmeng Zuo. Glad: Towards better recon- struction with global and local adaptive diffusion models for unsupervised anomaly detection. InProceedings of the Eu- ropean Conference on Computer Vision, pages 1–17, 2024. 1

2024
[38]

A unified model for multi-class anomaly detection

Zhiyuan You, Lei Cui, Yujun Shen, Kai Yang, Xin Lu, Yu Zheng, and Xinyi Le. A unified model for multi-class anomaly detection. InProceedings of the International Con- ference on Neural Information Processing Systems, pages 4571–4584, 2022. 3

2022
[39]

Fastflow: Unsupervised anomaly detection and localization via 2d normalizing flows.arXiv preprint arXiv:2111.07677, 2021

Jiawei Yu, Ye Zheng, Xiang Wang, Wei Li, Yushuang Wu, Rui Zhao, and Liwei Wu. Fastflow: Unsupervised anomaly detection and localization via 2d normalizing flows.arXiv preprint arXiv:2111.07677, 2021. 2, 3, 8

work page arXiv 2021
[40]

Wide Residual Networks

Sergey Zagoruyko. Wide residual networks.arXiv preprint arXiv:1605.07146, 2016. 6

work page internal anchor Pith review arXiv 2016
[41]

Dauphin, and David Lopez-Paz

Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimiza- tion. InProceedings of the International Conference on Learning Representations, 2018. 6

2018
[42]

Msflow: Multiscale flow-based framework for unsupervised anomaly detection.IEEE Transactions on Neural Networks and Learning Systems, pages 2437–2450,

Yixuan Zhou, Xing Xu, Jingkuan Song, Fumin Shen, and Heng Tao Shen. Msflow: Multiscale flow-based framework for unsupervised anomaly detection.IEEE Transactions on Neural Networks and Learning Systems, pages 2437–2450,