Pith. sign in

REVIEW 5 major objections 5 minor 39 references

Friend or Foe? Harnessing Controllable Overfitting for Anomaly Detection

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Deliberately overfitting an anomaly detector inside a monitored 'golden interval' improves detection: the paper's COAD framework raises AUROC on MVTec AD and VisA across RD, RD++, UniAD, and DiAD.

desk verdict The empirical overfitting recipe is worth a second look, but the ARQ control loop moves in the wrong direction and RADI is just AUROC. read the letter →

arxiv 2412.00560 v2 pith:ZQCH6JLS submitted 2024-11-30 cs.LG cs.AI

classification cs.LGcs.AI
keywords controllableoverfittinganomalydetectionAberranceRetentionQuotientRelativeDistributionIndexteacher-studentframeworkGaussiannoisepseudo-anomaliesDualControlMechanismAUROC
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 argues that overfitting, usually treated as a training failure to be avoided, can be deliberately switched on to make anomaly detectors more sensitive. It introduces COAD (Controllable Overfitting-based Anomaly Detection), a plug-in training stage that continues training an already-trained detector at one-tenth the learning rate while monitoring the drift between its predictions and the reference outputs it was trained to reproduce. The drift is measured by a new quantity, the Aberrance Retention Quotient (ARQ); its effect on detection is measured by another, the Relative Anomaly Distribution Index (RADI), defined as the probability that an anomalous pixel scores higher than a normal one. Under the assumptions that overfitting squeezes the normal-score variance exponentially while leaving the anomaly-score distribution alone, the paper derives a closed-form optimal overfitting level and a golden control interval, then reports image- and pixel-level AUROC gains on MVTec AD and VisA when COAD is appended to RD, RD++, UniAD, and DiAD. If the derivation is right, the payoff is a principled rule for deciding how long to overfit any reconstruction-based anomaly detector.

What carries the argument

The load-bearing machinery is the pair of metrics ARQ and RADI. ARQ, the Aberrance Retention Quotient, is defined as $ARQ = \sum_i |\hat{y}_i - y_i| \big/ \sum_i y_i$, where $y_i$ is the reference output (the teacher network's prediction in the teacher–student setting) and $\hat{y}_i$ is the model's prediction; it is the monitored dose of overfitting used to gate training. RADI, the Relative Anomaly Distribution Index, is the probability $P(S_a>S_n)$ that a random anomalous pixel scores above a random normal pixel, computed through the cumulative distribution functions of the two score populations. Under the paper's Gaussian assumptions the two metrics combine into one closed-form expression, $RADI(\theta)=\Phi\big((\mu_a-\mu_n)/\sqrt{\sigma_n(\theta)^2+\sigma_a^2}\big)$, and the exponential variance-decay model $\sigma_n(\theta)=\sigma_{n0}e^{-k\theta}+\sigma_{\mathrm{noise}}(\theta)$ turns maximizing detection performance into a one-variable calculus problem whose solution is the paper's Eq. (12). The Dual Control Mechanism operationalizes the theory: keep ARQ inside the golden interval and require $\frac{dRADI}{d\theta}\ge 0$; if either condition fails, progressively freeze the student's lower layers, preserving general feature extraction while letting higher layers keep their overfitted sensitivity.

What would settle it

Run any COAD-enhanced detector, for example RD++ on MVTec AD, and record ARQ together with the variances of normal-pixel and anomalous-pixel scores after every training step of the overfitting stage. If the normal variance does not decline monotonically as ARQ rises, or if the anomaly distribution drifts beyond sampling noise, Eqs. (4)–(5) fail and the derived optimum is not a peak. A second, assumption-free test sweeps ARQ across the claimed golden interval and computes empirical RADI from Eq. (6): if RADI does not peak inside the interval, or peaks outside it, the optimal-stopping rule is contradicted regardless of the parameter values.

Watch

Extended reading notes

Core claim

The paper's central claim is that the conventional ban on overfitting is wrong for anomaly detection: a network deliberately overfitted to normal training samples becomes better, not worse, at flagging anomalies, provided the overfitting is kept inside a monitored window. The mechanism is a distributional asymmetry. As the student memorizes normal patterns, its prediction errors on normal pixels become more concentrated—the paper models the normal-score standard deviation as $\sigma_n(\theta)=\sigma_{n0}e^{-k\theta}$, decaying exponentially with the overfitting level $\theta$—while the prediction errors on anomalous pixels, which the model never saw in training, keep their original spread. The separation between the two score populations is quantified by RADI $=P(S_a>S_n)$, which under Gaussian assumptions takes the closed form $\Phi\big((\mu_a-\mu_n)/\sqrt{\sigma_n(\theta)^2+\sigma_a^2}\big)$; setting its derivative to zero yields the optimal overfitting point $\theta^* = (\ln(k\sigma_{n0})-\ln(h\sigma_{\max})+h\theta_0)/(k-h)$, with a noise term $\sigma_{\mathrm{noise}}(\theta)$ marking the point where overfitting starts to damage the scores. Training is steered by a Dual Control Mechanism that keeps ARQ inside the golden interval $[\theta-\delta,\theta+\delta]$ and the RADI gradient non-negative, freezing the student's layers from the bottom up when both conditions fail. The paper also reports that anomalous-pixel score distributions sit close to Gaussian (total variation distance about 0.08), which it reads as a theoretical justification for training on Gaussian-noise pseudo-anomalies, and it argues that RADI is more informative than threshold-based AUROC because it models the separation of the two distributions directly.

Load-bearing premise

Everything rests on ARQ—the ratio measuring how far the model's predictions drift from the reference outputs it trains against, with no labels available—rising steadily as the model overfits, and on the distributional story that this drift squeezes the spread of normal-pixel scores exponentially (Eq. 4) while barely moving the spread of anomaly scores (Eq. 5).

Editorial extensions

If this is right

  • COAD becomes a drop-in final training stage: any teacher–student or reconstruction-based anomaly detector can be fine-tuned under the ARQ/RADI control loop, and the inference pipeline stays unchanged.
  • The overfitting level stops being an arbitrary choice: the golden interval supplies a monitoring rule, with reported operating points of ARQ 0.006 for one-class and 0.06 for multi-class tasks.
  • Gaussian noise gains a statistical justification as a pseudo-anomaly generator (TVD about 0.08), retroactively supporting the training recipes of existing detectors that synthesize anomalies this way.
  • Pixel-level localization improves on every tested framework, with the largest reported single gain an 8.7-point jump for RD on VisA.
  • The framework applies to both one-class and multi-class settings and to diffusion-based detectors, so the mechanism generalizes beyond the reverse-distillation family it was first demonstrated on.

Reading between the lines

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

  • A gap separates theory from practice in the paper: the optimal $\theta^*$ in Eq. (12) depends on constants ($k$, $h$, $\sigma_{\max}$, $\theta_0$) that are never estimated from data, and the operating points used in the experiments (ARQ 0.006 and 0.06) are chosen empirically, so the derivation currently predicts the shape of the performance curve rather than the exact best value.
  • The variance-shrinkage mechanism implies a directly measurable signature: during the overfitting stage the dispersion of normal-pixel scores should compress monotonically with rising ARQ while anomaly-pixel dispersion stays flat; logging these per training step on any public dataset would test Eqs. (4)–(5) without touching downstream AUROC.
  • RADI's claimed superiority over AUROC is testable in isolation: RADI is a single probability where AUROC integrates over thresholds, so score distributions exist for which the two order differently, and such a counterexample would delimit the claim.
  • The same logic—overfit a reference population that is abundant while treating rare deviations as anomalies—might transfer to non-industrial settings such as medical imaging or fraud detection, which the paper lists as future directions but does not evaluate.
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

5 major / 5 minor

Summary. The paper proposes COAD, a post-training procedure that continues to train student-teacher anomaly detectors after standard training, with the goal of deliberately overfitting inside a monitored 'golden interval' to sharpen anomaly scores. Overfitting is quantified by ARQ (Eq. 1), the normalized absolute deviation between student predictions and teacher features, and detection quality is measured by RADI (Eq. 6), defined as the probability that an anomalous score exceeds a normal score. The authors claim a closed-form optimal ARQ (Eq. 12), a dual control mechanism that freezes layers when ARQ leaves the golden interval or the RADI gradient turns negative, and a theoretical validation of Gaussian noise as pseudo-anomalies. Experiments on MVTec AD and VisA with RD, RD++, UniAD, and DiAD report AUROC gains. The central theoretical apparatus, however, contains internal inconsistencies and unverified claims, and the experiments do not provide the evidence needed to support the load-bearing assertions.

Significance. If the central claims were correct, the paper would introduce a practical and theoretically grounded way to exploit overfitting in anomaly detection, with potential value as a plug-in for existing student-teacher detectors. The experimental breadth, spanning four baselines and two datasets, is a genuine effort, and the authors are transparent about some limitations, notably that the ARQ values used in the experiments are chosen empirically. However, the significance is currently not established: RADI is definitionally equivalent to AUROC, the ARQ control signal has the wrong monotonic direction relative to its own definition, the closed-form optimum in Eq. (12) is algebraically inconsistent with Eqs. (8)-(11), and the experimental tables lack variance estimates. The conditional promise of the idea does not compensate for the fact that the paper's headline theoretical contributions do not survive scrutiny.

major comments (5)
  1. [Section 3.2, Eq. (1) and Algorithm 1] ARQ as defined in Eq. (1) is a normalized reconstruction error between student predictions and teacher features. Under the student's training objective of minimizing that reconstruction loss, overfitting makes the predictions approach the teacher features, so ARQ decreases rather than increases. Yet Eq. (4) assumes the normal-score standard deviation decays exponentially with increasing theta, and Algorithm 1 issues freeze commands when theta exceeds the optimal range. With the printed definitions, the freeze trigger would fire at the start of training, when reconstruction error is large, and would never fire during the overfitting stage the method is designed to exploit. This makes the control loop internally inconsistent and invalidates the golden-interval mechanism as stated.
  2. [Section 3.3.2, Eq. (6) and Section 3.3.3] RADI is defined as P(S_a > S_n) in Eq. (6). For two score distributions, this is exactly the definition of AUROC, equivalently the Mann-Whitney U statistic. The paper nevertheless claims RADI is 'superior to traditional AUROC' and treats it as a new metric, which is circular: any statement that RADI increases with overfitting is a statement that AUROC increases with overfitting. The text in Section 3.3.3 inadvertently confirms the identity by saying the derivative of 'AUROC_pixel(theta)' is being optimized. This undermines the claimed conceptual novelty and the theoretical motivation for the dual control mechanism.
  3. [Section 3.3.3, Eq. (12) and Section 4.1] Solving d sigma_n(theta)/d theta = 0 with Eqs. (8) and (9) yields theta* = [ln(k sigma_n0) - ln(h sigma_max) - h theta0]/(k - h), not the printed expression with '+ h theta0'. The sign error makes Eq. (12) inconsistent with the preceding derivation. Moreover, Eq. (12) is never evaluated in the experiments; Section 4.1 explicitly states that the ARQ values 0.006 and 0.06 are 'chosen empirically'. The claimed theoretical derivation of the golden interval is therefore not exercised, and the central claim that COAD provides a principled, theoretically grounded way to control overfitting is unsupported by the reported evidence.
  4. [Section 4.2.1, Tables 4 and 5] The reported AUROC values are single numbers with no standard deviations, no number of seeds, and no significance tests. Several reported gains are small (0.3-1.2 AUROC), and one ablation shows a regression on VisA image-level AUROC for RD (96.0 to 95.9). Without error bars or repeated runs, the claim of 'consistent and notable improvements' is not established. This is a load-bearing issue because the entire empirical case for COAD rests on these small deltas over baseline frameworks.
  5. [Section 4.3 and Appendix C] The Gaussian-noise validation does not directly compare the distribution of injected Gaussian noise with the distribution of real anomalies. Instead, Table 8 and Figures 5-7 report Total Variation Distance between model prediction-score histograms and fitted Gaussian curves, which validates a normality assumption about scores, not the suitability of Gaussian noise as a pseudo-anomaly generator in training. The claim that Gaussian noise is statistically validated as a pseudo-anomaly source therefore goes beyond what the presented evidence supports. A direct controlled comparison with alternative noise types or pseudo-anomaly generators is needed.
minor comments (5)
  1. [Section 3.3.1, Eq. (4)] The text calls sigma_n0 the 'initial variance', but Eq. (4) defines sigma_n(theta) as a standard deviation; the terminology should be corrected to avoid confusion.
  2. [Algorithm 1, line 22 and Eq. (2)] The condition 'theta exceeds theta_optimal' is ambiguous because Eq. (2) defines ARQ_optimal as an interval [theta-delta, theta+delta]; the algorithm should specify that the condition is exceeding the upper bound theta+delta.
  3. [Appendix C.3] The acronym ARQ is expanded as 'Anomaly Rate Quotient' rather than 'Aberrance Retention Quotient' as used in the main text; this inconsistency should be fixed.
  4. [Algorithm 3 and Figure 2] Algorithm 3 refers to a filtered feature input and an optional pseudo-anomaly filter, but no such filter is defined in the method section or caption, leaving the inference procedure underspecified.
  5. [Section 4.2.1] The text states that the ablation shows 'consistent and notable improvements', but the VisA RD image-level AUROC decreases from 96.0 to 95.9; the authors should acknowledge and discuss this counterexample rather than claiming consistency without qualification.

Circularity Check

2 steps flagged · score 7.0 of 10

RADI is a relabeled AUROC, and the golden overfitting interval is selected post hoc from the benchmark tables rather than predicted from Eq. 12; the central theoretical control claims reduce to definition and empirical fitting.

  1. self definitional [Sec. 3.3.2, Eq. (6); Sec. 3.3.3, text before Eq. (10)]
    "RADI(ARQ) = P(S_a > S_n) = ∫ P(S_a > x) f_Sn(x) dx, (6) ... To find the optimal Aberrance Retention Quotient of ARQ = θ* that maximizes RADI, we calculate the derivative ofAUROCpixel(θ) and set it to zero."

    For two score distributions, P(S_a > S_n) is exactly the AUROC (equivalently, the Mann-Whitney/Wilcoxon statistic). The paper introduces RADI as "a metric superior to traditional AUROC" and calls it a complementary metric to AUROC-pixel, then optimizes "the derivative of AUROC_pixel" to maximize RADI. Thus RADI is not an independent derived quantity; the claimed theoretical bridge between overfitting and detection performance is just the outcome metric AUROC relabeled and re-derived.

  2. fitted input called prediction [Sec. 4.1, paragraph after Eq. (14); Sec. 4.2.2, Tables 6 and 7]
    "Using the optimal ARQ value θ derived from the ARQoptimal as indicated by eq. (12), we set ARQ to 0.006 for one-class tasks and 0.06 for multi-class tasks as shown in section 4.2.2 which is chosen empirically. ARQoptimal is defined as θ ∈ [0.001, 0.011] for one-class tasks and θ ∈ [0.01, 0.11] for multi-class tasks, where the balance between overfitting and generalization is optimal for anomaly detection."

    The formula in Eq. (12) is never evaluated to produce the operating points: the paper states the values "which is chosen empirically." The "golden interval" is instead read off Tables 6 and 7, which compare AUROC under distinct ARQ ranges on the same test benchmarks. Selecting the range that achieved the best pixel-level AUROC and then calling it ARQoptimal means the predicted optimum is a post hoc fit to the outcome, so the Dual Control Mechanism in Eq. (14) is keyed to a retrospectively chosen interval rather than to a derived prediction.

full rationale

The empirical contribution—running RD/RD++/UniAD/DiAD with and without extra overfitting training and reporting benchmark AUROC—is an independent measurement and is not itself circular. However, the paper's two central theoretical advances are. First, RADI is defined as P(S_a>S_n), which is the definition of AUROC, and the paper even optimizes "the derivative of AUROC_pixel" when "maximizing RADI"; the claimed "metric superior to traditional AUROC" is therefore AUROC under a new name. Second, the "optimal ARQ" claimed to follow from Eq. (12) is explicitly "chosen empirically," and the golden interval coincides with the best-performing rows of the ablation tables, so the theoretical freeze/control law is not exercised as a prediction. These are not self-citation issues (I found no load-bearing self-citations in the reference list), and the empirical SOTA claims retain independent content, so a mid-high score rather than 10 is appropriate. A separate, non-circular correctness concern is that Eq. (1) makes ARQ decrease as student-teacher reconstruction error drops, while Eq. (4) and Algorithm 1 treat ARQ as increasing with overfitting; I list this as a consistency flaw, not as circularity.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The framework's theoretical core depends on six assumed distributional and functional forms with no estimated constants; the only quantities actually set, theta and delta, are chosen empirically from validation results. No new physical entities are introduced.

free parameters (8)
  • theta (target ARQ) = 0.006 (one-class), 0.06 (multi-class)
    Set empirically in Sec. 4.1; Eq. 12 is not evaluated because k, h, sigma_max, and theta_0 are unmeasured.
  • delta (ARQ interval half-width) = 0.005 / 0.05 implied by intervals [0.001, 0.011] and [0.01, 0.11]
    Defined as 'same order of magnitude' and selected through validation search in Sec. 4.2.2.
  • k (variance decay rate) = not specified
    Exponential decay constant in Eq. 4, assumed without estimation.
  • sigma_n0 (initial normal score standard deviation) = not specified
    Initial variance in Eq. 4, never estimated.
  • sigma_max (noise saturation) = not specified
    Maximum noise standard deviation in Eq. 9, never estimated.
  • h (noise onset rate) = not specified
    Rate constant in Eq. 9, never estimated.
  • theta_0 (noise onset ARQ) = not specified
    Threshold in Eq. 9, never estimated.
  • C_thr (freeze counter threshold) = not specified
    Threshold in Algorithm 2 for deciding when to freeze a layer; no value is given.
assumptions (6)
  • domain assumption Normal pixel prediction scores are Gaussian with mean and variance depending on ARQ (Eq. 3).
    Needed for the closed-form RADI in Eq. 7; only visual histogram fits in Sec. 4.3 are offered as support.
  • domain assumption Anomalous pixel prediction scores are Gaussian and independent of ARQ (Eq. 5).
    Assumed so RADI(theta) changes only through sigma_n(theta); no mechanism justifies stability under overfitting.
  • ad hoc to paper Normal-score standard deviation decays exponentially with ARQ (Eq. 4).
    The functional form is asserted, not derived; the later optimum depends on this exact form.
  • ad hoc to paper Overfitting noise term sigma_noise(theta) has the saturating form of Eq. 9.
    Introduced to create an interior maximum; parameters are unmeasured.
  • domain assumption ARQ can be computed on unsupervised training data using 'ground truth values' y_i.
    Eq. 1 uses y_i but no labels exist in the one-class setting; the operational definition is not specified.
  • standard math Central limit theorem justifies Gaussian score distributions.
    Invoked in Sec. 3.3.1; acceptable as background, but it does not fix the functional dependence on ARQ.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Friend or Foe? Harnessing Controllable Overfitting for Anomaly Detection." pith.science (2026). https://pith.science/paper/ZQCH6JLS

@misc{pith2026241200560,
  author       = {Pith},
  title        = {Pith review of: Friend or Foe? Harnessing Controllable Overfitting for Anomaly Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZQCH6JLS}},
  note         = {Machine review of arXiv:2412.00560}
}
read the original abstract

Overfitting has traditionally been viewed as detrimental to anomaly detection, where excessive generalization often limits models' sensitivity to subtle anomalies. Our work challenges this conventional view by introducing Controllable Overfitting-based Anomaly Detection (COAD), a novel framework that strategically leverages overfitting to enhance anomaly discrimination capabilities. We propose the Aberrance Retention Quotient (ARQ), a novel metric that systematically quantifies the extent of overfitting, enabling the identification of an optimal golden overfitting interval wherein model sensitivity to anomalies is maximized without sacrificing generalization. To comprehensively capture how overfitting affects detection performance, we further propose the Relative Anomaly Distribution Index (RADI), a metric superior to traditional AUROC by explicitly modeling the separation between normal and anomalous score distributions. Theoretically, RADI leverages ARQ to track and evaluate how overfitting impacts anomaly detection, offering an integrated approach to understanding the relationship between overfitting dynamics and model efficacy. We also rigorously validate the statistical efficacy of Gaussian noise as pseudo-anomaly generators, reinforcing the method's broad applicability. Empirical evaluations demonstrate that our controllable overfitting method achieves State-Of-The-Art(SOTA) performance in both one-class and multi-class anomaly detection tasks, thus redefining overfitting as a powerful strategy rather than a limitation.

Figures

Figures reproduced from arXiv: 2412.00560 by the authors.

Figure 1
Figure 1. Each subfigure presents a sequence from left to [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comprehensive Framework for COAD. In the Traditional Model, Teacher Model remains frozen throughout. The pseudo-anomaly Gaussian Noise is optional. The student model also remains unfrozen during normal training stage, with selective freezing employed during the controllable overfitting stages. When the Dual Control Mechanism fails, Freeze Command Signals are triggered, prompting the selective freezing of certain net… view at source ↗
Figure 3
Figure 3. Qualitative visualization results on partial cate [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Qualitative visualization results on partial cate [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Histograms for normal and anomaly pixel scores [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Histograms for anomaly scores of 15 individual categories. Each histogram includes Gaussian fitting curves for better [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Histograms for normal scores of 15 individual categories. Each histogram includes Gaussian fitting curves for better [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Qualitative Comparison of Anomaly Detection Results Across 15 Categories Using the Baseline Method [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: Qualitative Comparison of Anomaly Detection Results Across 15 Categories Using a Baseline Method [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Qualitative Comparison of Anomaly Detection Results Across 15 Categories Using a Baseline Method [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Qualitative Comparison of Anomaly Detection Results Across 15 Categories Using a Baseline Method [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 25 canonical work pages

  1. [1]

    Paul Bergmann, Michael Fauser, David Sattlegger, and Carsten Steger. 2019. MVTec AD–A comprehensive real-world dataset for unsupervised anomaly de- tection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 9592–9600

  2. [2]

    Paul Bergmann, Michael Fauser, David Sattlegger, and Carsten Steger. 2020. Uninformed Students: Student-Teacher Anomaly Detection With Discriminative Latent Embeddings. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

  3. [3]

    Christopher Bishop. 2006. Pattern Recognition and Machine Learning . Springer. https://www.microsoft.com/en-us/research/publication/pattern- recognition-machine-learning/

  4. [4]

    Qiyu Chen, Huiyuan Luo, Chengkan Lv, and Zhengtao Zhang. 2024. A Unified Anomaly Synthesis Strategy with Gradient Ascent for Industrial Anomaly Detec- tion and Localization. arXiv:2407.09359 [cs.CV] https://arxiv.org/abs/2407.09359

  5. [5]

    Thomas Defard, Aleksandr Setkov, Angelique Loesch, and Romaric Audigier. 2021. PaDiM: A Patch Distribution Modeling Framework for Anomaly Detection and Localization. In Pattern Recognition. ICPR International Workshops and Challenges, Alberto Del Bimbo, Rita Cucchiara, Stan Sclaroff, Giovanni Maria Farinella, Tao Mei, Marco Bertini, Hugo Jair Escalante, a...

  6. [6]

    Hanqiu Deng and Xingyu Li. 2022. Anomaly Detection via Reverse Distilla- tion From One-Class Embedding. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 9737–9746

  7. [7]

    Matic Fučka, Vitjan Zavrtanik, and Danijel Skočaj. 2024. TransFusion – A Transparency-Based Diffusion Model for Anomaly Detection. In Proc. of ECCV (35). 91–108

  8. [8]

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. 2016. Deep Learning. MIT Press. http://www.deeplearningbook.org

Show all 39 references
  1. [9]

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Y. Bengio. 2014. Generative Adversarial Nets. MIT Press (2014)

  2. [10]

    Denis Gudovskiy, Shun Ishizaka, and Kazuki Kozuka. 2021. CFLOW-AD: Real- Time Unsupervised Anomaly Detection with Localization via Conditional Nor- malizing Flows. arXiv:2107.12571 [cs.CV] https://arxiv.org/abs/2107.12571

  3. [11]

    Trevor Hastie, Robert Tibshirani, Jerome Friedman, and James Franklin. 2004. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Math. Intell. 27 (11 2004), 83–85. doi:10.1007/BF02985802

  4. [12]

    Haoyang He, Jiangning Zhang, Hongxu Chen, Xuhai Chen, Zhishan Li, Xu Chen, Yabiao Wang, Chengjie Wang, and Lei Xie. 2023. DiAD: A Diffusion- based Framework for Multi-class Anomaly Detection. arXiv:2312.06607 [cs.CV] https://arxiv.org/abs/2312.06607

  5. [13]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33 (2020), 6840–6851

  6. [14]

    Teng Hu, Jiangning Zhang, Ran Yi, Yuzhen Du, Xu Chen, Liang Liu, Yabiao Wang, and Chengjie Wang. 2024. AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model. In Proc. of AAAI, Vol. 38. 8526–8534

  7. [15]

    Jiarui Lei, Xiaobo Hu, Yue Wang, and Dong Liu. 2023. PyramidFlow: High- Resolution Defect Contrastive Localization using Pyramid Normalizing Flow. arXiv:2303.02595 [cs.CV] https://arxiv.org/abs/2303.02595

  8. [16]

    Chun Liang Li, Kihyuk Sohn, Jinsung Yoon, and Tomas Pfister. 2021. CutPaste: Self-Supervised Learning for Anomaly Detection and Localization

  9. [17]

    Yufei Liang, Jiangning Zhang, Shiwei Zhao, Runze Wu, Yong Liu, and Shuwen Pan. 2023. Omni-Frequency Channel-Selection Representations for Unsupervised Anomaly Detection. IEEE Transactions on Image Processing 32 (2023), 4327–4340. doi:10.1109/TIP.2023.3293772

  10. [18]

    Xinyue Liu, Jianyuan Wang, Biao Leng, and Shuo Zhang. 2024. Dual-Modeling Decouple Distillation for Unsupervised Anomaly Detection. arXiv preprint arXiv:2408.03888 (2024)

  11. [19]

    Zhikang Liu, Yiming Zhou, Yuansheng Xu, and Zilei Wang. 2023. SimpleNet: A Simple Network for Image Anomaly Detection and Localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 20402–20411

  12. [20]

    Ruiying Lu, YuJie Wu, Long Tian, Dongsheng Wang, Bo Chen, Xiyang Liu, and Ruimin Hu. 2023. Hierarchical vector quantized transformer for multi-class unsupervised anomaly detection. Advances in Neural Information Processing Systems 36 (2023), 8487–8500

  13. [21]

    Henry B Mann and Donald R Whitney. 1947. On a test of whether one of two random variables is stochastically larger than the other. The annals of mathematical statistics (1947), 50–60

  14. [22]

    Kevin P Murphy. 2012. Machine learning: a probabilistic perspective . Cambridge, MA

  15. [23]

    Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Schölkopf, Thomas Brox, and Peter Gehler. 2022. Towards Total Recall in Industrial Anomaly Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 14318–14328. Conference’17, J...

  16. [24]

    Rohban, and Hamid R

    Mohammadreza Salehi, Niousha Sadjadi, Soroosh Baselizadeh, Mohammad H. Rohban, and Hamid R. Rabiee. 2021. Multiresolution Knowledge Distillation for Anomaly Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 14902–14912

  17. [25]

    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. 2023. Revisiting Reverse Distillation for Anomaly Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVP...

  18. [26]

    Frank Wilcoxon. 1992. Individual comparisons by ranking methods. In Break- throughs in statistics: Methodology and distribution . Springer, 196–202

  19. [27]

    Feng Xiao, Ruoyu Sun, and Jicong Fan. 2023. Restricted Generative Projection for One-Class Classification and Anomaly Detection. arXiv:2307.04097 [cs.LG] https://arxiv.org/abs/2307.04097

  20. [28]

    Hang Yao, Ming Liu, Haolin Wang, Zhicun Yin, Zifei Yan, Xiaopeng Hong, and Wangmeng Zuo. 2024. GLAD: Towards Better Reconstruction with Global and Local Adaptive Diffusion Models for Unsupervised Anomaly Detection. arXiv preprint arXiv:2406.07487 (2024)

  21. [29]

    Xincheng Yao, Ruoqi Li, Jing Zhang, Jun Sun, and Chongyang Zhang. 2023. Explicit Boundary Guided Semi-Push-Pull Contrastive Learning for Supervised Anomaly Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 24490–24499

  22. [30]

    Jihun Yi and Sungroh Yoon. 2020. Patch SVDD: Patch-level SVDD for Anomaly Detection and Segmentation. In Proceedings of the Asian Conference on Computer Vision (ACCV)

  23. [31]

    Zhiyuan You, Lei Cui, Yujun Shen, Kai Yang, Xin Lu, Yu Zheng, and Xinyi Le. 2022. A Unified Model for Multi-class Anomaly Detection. arXiv:2206.03687 [cs.CV] https://arxiv.org/abs/2206.03687

  24. [32]

    Danijel Zavrtanik, Matej Skocaj. 2021. Reconstruction by inpainting for visual anomaly detection. Pattern Recognition: The Journal of the Pattern Recognition Society 112, 1 (2021)

  25. [33]

    Vitjan Zavrtanik, Matej Kristan, and Danijel Skočaj. 2022. DSR – A Dual Subspace Re-Projection Network for Surface Anomaly Detection. In Computer Vision – ECCV 2022, Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, and Tal Hassner (Eds.). Springer Natur...

  26. [34]

    Vitjan Zavrtanik, Matej Kristan, and Danijel Skočaj. 2021. DRAEM - A Discrimi- natively Trained Reconstruction Embedding for Surface Anomaly Detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . 8330–8339

  27. [35]

    Ximiao Zhang, Min Xu, and Xiuzhuang Zhou. 2024. RealNet: A Feature Se- lection Network with Realistic Synthetic Anomaly for Anomaly Detection. arXiv:2403.05897 [cs.CV] https://arxiv.org/abs/2403.05897

  28. [36]

    Ying Zhao. 2023. OmniAL: A Unified CNN Framework for Unsupervised Anomaly Localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 3924–3933

  29. [37]

    Yibo Zhou. 2022. Rethinking Reconstruction Autoencoder-Based Out-of- Distribution Detection. (2022)

  30. [38]

    Yang Zou, Jongheon Jeong, Latha Pemula, Dongqing Zhang, and Onkar Dabeer

  31. [2022]

    arXiv preprint arXiv:2207.14315 (2022)

    SPot-the-Difference Self-Supervised Pre-training for Anomaly Detection and Segmentation. arXiv preprint arXiv:2207.14315 (2022). Friend or Foe? Harnessing Controllable Overfitting for Anomaly Detection Conference’17, July 2017, Washington, DC, USA Appendix A Dual Control Mecha...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.