pith. sign in

arxiv: 2502.20981 · v3 · pith:DN6FUS5Fnew · submitted 2025-02-28 · 💻 cs.CV

Distribution Prototype Diffusion Learning for Open-set Supervised Anomaly Detection

Fuyun Wang , Tong Zhang , Yuanzhi Wang , Yide Qiu , Xin Liu , Xu Guo , Zhen Cui This is my paper

Pith reviewed 2026-05-23 02:09 UTC · model grok-4.3

classification 💻 cs.CV
keywords open-set supervised anomaly detectiondistribution prototype diffusionGaussian prototypesSchrödinger bridgehyperspherical dispersionanomaly boundary learninglatent representation space
0
0 comments X

The pith

DPDL uses learnable Gaussian prototypes and a Schrödinger bridge to enclose normal samples in a compact discriminative space for open-set anomaly detection.

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

The paper aims to improve open-set supervised anomaly detection by focusing on normal sample priors rather than generating pseudo anomalies. It constructs multiple learnable Gaussian prototypes to represent normal data in latent space and applies a Schrödinger bridge diffusion process that moves normal samples toward these prototypes while directing anomalies away. Dispersion learning in hyperspherical space further aids separation of out-of-distribution samples. This yields state-of-the-art results on nine public datasets without requiring post-hoc tuning or dataset-specific adjustments.

Core claim

The central claim is that multiple learnable Gaussian prototypes create a latent representation space for diverse normal samples, and learning a Schrödinger bridge enables diffusive transitions that pull normal samples toward the prototypes while steering anomalies away, with added hyperspherical dispersion learning to enhance inter-sample separation and produce reliable boundaries for detecting unseen anomalies.

What carries the argument

Multiple learnable Gaussian prototypes paired with a Schrödinger bridge diffusion process that guides normal samples toward the prototypes and anomalies away from them, plus dispersion feature learning in hyperspherical space.

If this is right

  • Normal samples gain a more abundant and diverse latent representation through the Gaussian prototypes.
  • Anomaly samples are actively steered away from the normal distribution space during the diffusion process.
  • Hyperspherical dispersion features improve identification of out-of-distribution anomalies.
  • The overall approach achieves state-of-the-art detection on nine public datasets without post-hoc tuning.

Where Pith is reading between the lines

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

  • The prototype-diffusion idea could be adapted to other settings where abundant normal data must be distinguished from rare or novel outliers.
  • Replacing pseudo-anomaly generation with prototype-based diffusion might simplify training pipelines in related detection tasks.
  • Testing the method on streaming or real-time data would reveal whether the learned boundaries remain stable over time.

Load-bearing premise

That learnable Gaussian prototypes and a Schrödinger bridge diffusion process will automatically form a compact boundary around normal samples that separates unseen anomalies without needing extra tuning or adjustments for each dataset.

What would settle it

Evaluate the method on a new dataset containing anomalies from distributions absent in training and check whether detection performance falls below existing methods or requires dataset-specific hyperparameter changes.

Figures

Figures reproduced from arXiv: 2502.20981 by Fuyun Wang, Tong Zhang, Xin Liu, Xu Guo, Yide Qiu, Yuanzhi Wang, Zhen Cui.

Figure 1
Figure 1. Figure 1: Our proposed DPDL framework. It comprises three distinct modules: Distribution Prototype Learning (DPL, Sec. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Ablation study for SB and DFL under the general settings and hard settings. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Parameter sensitivity analysis for C, ϵ, κ and λ [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

In Open-set Supervised Anomaly Detection (OSAD), the existing methods typically generate pseudo anomalies to compensate for the scarcity of observed anomaly samples, while overlooking critical priors of normal samples, leading to less effective discriminative boundaries. To address this issue, we propose a Distribution Prototype Diffusion Learning (DPDL) method aimed at enclosing normal samples within a compact and discriminative distribution space. Specifically, we construct multiple learnable Gaussian prototypes to create a latent representation space for abundant and diverse normal samples and learn a Schr\"odinger bridge to facilitate a diffusive transition toward these prototypes for normal samples while steering anomaly samples away. Moreover, to enhance inter-sample separation, we design a dispersion feature learning way in hyperspherical space, which benefits the identification of out-of-distribution anomalies. Experimental results demonstrate the effectiveness and superiority of our proposed DPDL, achieving state-of-the-art performance on 9 public datasets.

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.

Referee Report

0 major / 2 minor

Summary. The paper proposes Distribution Prototype Diffusion Learning (DPDL) for open-set supervised anomaly detection (OSAD). It constructs multiple learnable Gaussian prototypes to model a compact latent space for normal samples, learns a Schrödinger bridge diffusion process that transitions normal samples toward the prototypes while repelling anomalies, and adds hyperspherical dispersion feature learning to improve inter-sample separation. The central empirical claim is that this combination yields state-of-the-art performance on nine public datasets with hyperparameters fixed across datasets and no post-hoc tuning.

Significance. If the reported results hold, the work offers a concrete alternative to pseudo-anomaly generation by directly encoding normal-sample priors via prototypes and a diffusion bridge; the fixed-hyperparameter regime across datasets would be a practical strength for generalization claims in OSAD.

minor comments (2)
  1. Abstract: the SOTA claim is stated without any quantitative deltas, dataset names, or baseline references, which is atypical even for an abstract and forces readers to reach the experimental section for any assessment of the central claim.
  2. The manuscript would benefit from an explicit statement (perhaps in §3 or §4) confirming that the number and initialization of Gaussian prototypes are the only free parameters and that all other quantities are derived without additional dataset-specific fitting.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their careful reading, positive summary of our contributions, and recommendation of minor revision. The report does not enumerate any specific major comments, so we have no individual points to address at this time.

Circularity Check

0 steps flagged

No significant circularity; new architectural components are independently defined

full rationale

The paper introduces DPDL via explicitly defined components (learnable Gaussian prototypes, Schrödinger bridge diffusion, hyperspherical dispersion) whose loss formulations and training procedure do not reduce by construction to quantities already fitted from prior data or self-citations. The central claim of compact normal enclosure is presented as an empirical outcome of supervised training on the nine datasets, with no load-bearing step that renames a fitted parameter as a prediction or imports uniqueness solely from overlapping-author prior work. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on the modeling choice of multiple Gaussian prototypes as targets for normal data and the applicability of Schrödinger bridge diffusion in this latent space; no explicit free parameters or invented entities are quantified in the abstract.

free parameters (1)
  • Number and initialization of Gaussian prototypes
    Learnable prototypes are introduced as a core modeling device whose count and starting values are not specified in the abstract.
axioms (1)
  • domain assumption Normal samples admit a compact multi-Gaussian representation in the learned latent space that separates them from anomalies
    This premise underpins the construction of prototypes and the diffusion objective.

pith-pipeline@v0.9.0 · 5695 in / 1151 out tokens · 77203 ms · 2026-05-23T02:09:12.461260+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · 1 internal anchor

  1. [1]

    Ub- normal: New benchmark for supervised open-set video anomaly detection

    Andra Acsintoae, Andrei Florescu, Mariana-Iuliana Georgescu, Tudor Mare, Paul Sumedrea, Radu Tudor Ionescu, Fahad Shahbaz Khan, and Mubarak Shah. Ub- normal: New benchmark for supervised open-set video anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 20143–20153, 2022. 1, 2

    work page 2022

  2. [2]

    Supervised anomaly detection for complex indus- trial images

    Aimira Baitieva, David Hurych, Victor Besnier, and Olivier Bernard. Supervised anomaly detection for complex indus- trial images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17754– 17762, 2024. 1

    work page 2024

  3. [3]

    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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9592–9600, 2019. 6, 1

    work page 2019

  4. [4]

    Hyperkvasir, a comprehensive multi-class im- age and video dataset for gastrointestinal endoscopy

    Hanna Borgli, Vajira Thambawita, Pia H Smedsrud, Steven Hicks, Debesh Jha, Sigrun L Eskeland, Kristin Ranheim Randel, Konstantin Pogorelov, Mathias Lux, Duc Tien Dang Nguyen, et al. Hyperkvasir, a comprehensive multi-class im- age and video dataset for gastrointestinal endoscopy. Scien- tific data, 7(1):283, 2020. 6, 1

    work page 2020

  5. [5]

    On the relation between optimal transport and schr ¨odinger bridges: A stochastic control viewpoint

    Yongxin Chen, Tryphon T Georgiou, and Michele Pavon. On the relation between optimal transport and schr ¨odinger bridges: A stochastic control viewpoint. Journal of Opti- mization Theory and Applications, 169:671–691, 2016. 2

    work page 2016

  6. [6]

    Generating and reweighting dense contrastive pat- terns for unsupervised anomaly detection

    Songmin Dai, Yifan Wu, Xiaoqiang Li, and Xiangyang Xue. Generating and reweighting dense contrastive pat- terns for unsupervised anomaly detection. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 1454– 1462, 2024. 1

    work page 2024

  7. [7]

    Diffusion schr¨odinger bridge with applications to score-based generative modeling

    Valentin De Bortoli, James Thornton, Jeremy Heng, and Ar- naud Doucet. Diffusion schr¨odinger bridge with applications to score-based generative modeling. Advances in Neural In- formation Processing Systems, 34:17695–17709, 2021. 2

    work page 2021

  8. [8]

    Automatic classification of defective photovoltaic module cells in electroluminescence images

    Sergiu Deitsch, Vincent Christlein, Stephan Berger, Claudia Buerhop-Lutz, Andreas Maier, Florian Gallwitz, and Chris- tian Riess. Automatic classification of defective photovoltaic module cells in electroluminescence images. Solar Energy, 185:455–468, 2019. 6, 1

    work page 2019

  9. [9]

    Catching both gray and black swans: Open-set supervised anomaly detection

    Choubo Ding, Guansong Pang, and Chunhua Shen. Catching both gray and black swans: Open-set supervised anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7388–7398,

    work page

  10. [10]

    Light and optimal schr ¨odinger bridge matching

    Nikita Gushchin, Sergei Kholkin, Evgeny Burnaev, and Alexander Korotin. Light and optimal schr ¨odinger bridge matching. In Forty-first International Conference on Ma- chine Learning, 2024. 2, 4

    work page 2024

  11. [11]

    Learning unified reference rep- resentation for unsupervised multi-class anomaly detection

    Liren He, Zhengkai Jiang, Jinlong Peng, Liang Liu, Qian- gang Du, Xiaobin Hu, Wenbing Zhu, Mingmin Chi, Yabiao Wang, and Chengjie Wang. Learning unified reference rep- resentation for unsupervised multi-class anomaly detection. arXiv preprint arXiv:2403.11561, 2024. 1

    work page arXiv 2024

  12. [12]

    Anomalyd- iffusion: Few-shot anomaly image generation with diffusion model

    Teng Hu, Jiangning Zhang, Ran Yi, Yuzhen Du, Xu Chen, Liang Liu, Yabiao Wang, and Chengjie Wang. Anomalyd- iffusion: Few-shot anomaly image generation with diffusion model. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 8526–8534, 2024. 1

    work page 2024

  13. [13]

    Comparison of novelty detection methods for multispectral images in rover-based planetary exploration missions

    Hannah R Kerner, Kiri L Wagstaff, Brian D Bue, Danika F Wellington, Samantha Jacob, Paul Horton, James F Bell, Chiman Kwan, and Heni Ben Amor. Comparison of novelty detection methods for multispectral images in rover-based planetary exploration missions. Data Mining and Knowledge Discovery, 34:1642–1675, 2020. 6, 1

    work page 2020

  14. [14]

    Unpaired image-to-image translation via neu- ral schr\” odinger bridge

    Beomsu Kim, Gihyun Kwon, Kwanyoung Kim, and Jong Chul Ye. Unpaired image-to-image translation via neu- ral schr\” odinger bridge. arXiv preprint arXiv:2305.15086,

    work page arXiv

  15. [15]

    San- flow: Semantic-aware normalizing flow for anomaly detec- tion

    Daehyun Kim, Sungyong Baik, and Tae Hyun Kim. San- flow: Semantic-aware normalizing flow for anomaly detec- tion. Advances in Neural Information Processing Systems , 36:75434–75454, 2023. 1

    work page 2023

  16. [16]

    Fast ensem- bling with diffusion schr \” odinger bridge

    Hyunsu Kim, Jongmin Yoon, and Juho Lee. Fast ensem- bling with diffusion schr \” odinger bridge. arXiv preprint arXiv:2404.15814, 2024. 2

    work page arXiv 2024

  17. [17]

    Light Schrödinger bridge.arXiv preprint arXiv:2310.01174, 2023

    Alexander Korotin, Nikita Gushchin, and Evgeny Bur- naev. Light schr \” odinger bridge. arXiv preprint arXiv:2310.01174, 2023. 2, 4

    work page arXiv 2023

  18. [18]

    A survey of the Schrödinger problem and some of its connections with optimal transport.arXiv preprint arXiv:1308.0215, 2013

    Christian L ´eonard. A survey of the schr \” odinger problem and some of its connections with optimal transport. arXiv preprint arXiv:1308.0215, 2013. 2, 3

    work page arXiv 2013

  19. [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. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 9664–9674, 2021. 1, 6

    work page 2021

  20. [20]

    Efficient anomaly detection with budget anno- tation using semi-supervised residual transformer

    Hanxi Li, Jingqi Wu, Hao Chen, Mingwen Wang, and Chun- hua Shen. Efficient anomaly detection with budget anno- tation using semi-supervised residual transformer. arXiv preprint arXiv:2306.03492, 2023. 1

    work page arXiv 2023

  21. [21]

    Promptad: Learn- ing prompts with only normal samples for few-shot anomaly detection

    Xiaofan Li, Zhizhong Zhang, Xin Tan, Chengwei Chen, Yanyun Qu, Yuan Xie, and Lizhuang Ma. Promptad: Learn- ing prompts with only normal samples for few-shot anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16838– 16848, 2024. 1

    work page 2024

  22. [22]

    Coft-ad: Contrastive fine-tuning for few-shot anomaly detection

    Jingyi Liao, Xun Xu, Manh Cuong Nguyen, Adam Goodge, and Chuan Sheng Foo. Coft-ad: Contrastive fine-tuning for few-shot anomaly detection. IEEE Transactions on Image Processing, 2024. 1

    work page 2024

  23. [23]

    Deep generalized schr ¨odinger bridge

    Guan-Horng Liu, Tianrong Chen, Oswin So, and Evangelos Theodorou. Deep generalized schr ¨odinger bridge. Advances in Neural Information Processing Systems , 35:9374–9388,

    work page

  24. [24]

    Gen- eralized schr \” odinger bridge matching

    Guan-Horng Liu, Yaron Lipman, Maximilian Nickel, Brian Karrer, Evangelos A Theodorou, and Ricky TQ Chen. Gen- eralized schr \” odinger bridge matching. arXiv preprint arXiv:2310.02233, 2023. 2

    work page arXiv 2023

  25. [25]

    i2- sb: Image-to-image schr \” odinger bridge

    Guan-Horng Liu, Arash Vahdat, De-An Huang, Evange- los A Theodorou, Weili Nie, and Anima Anandkumar. i2- sb: Image-to-image schr \” odinger bridge. arXiv preprint arXiv:2302.05872, 2023. 2

    work page arXiv 2023

  26. [26]

    Unsupervised continual anomaly detection with contrastively-learned prompt

    Jiaqi Liu, Kai Wu, Qiang Nie, Ying Chen, Bin-Bin Gao, Yong Liu, Jinbao Wang, Chengjie Wang, and Feng Zheng. Unsupervised continual anomaly detection with contrastively-learned prompt. In Proceedings of the AAAI Conference on Artificial Intelligence , pages 3639–3647,

    work page

  27. [27]

    Margin learning embedded prediction for video anomaly detection with a few anomalies

    Wen Liu, Weixin Luo, Zhengxin Li, Peilin Zhao, Shenghua Gao, et al. Margin learning embedded prediction for video anomaly detection with a few anomalies. In IJCAI, pages 023–3, 2019. 1, 2, 6

    work page 2019

  28. [28]

    Dual-modeling decouple distillation for unsuper- vised anomaly detection

    Xinyue Liu, Jianyuan Wang, Biao Leng, and Shuo Zhang. Dual-modeling decouple distillation for unsuper- vised anomaly detection. arXiv preprint arXiv:2408.03888,

    work page arXiv

  29. [29]

    Decoupled Weight Decay Regularization

    I Loshchilov. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  30. [30]

    Directional statistics

    Kanti V Mardia and Peter E Jupp. Directional statistics. John Wiley & Sons, 2009. 5

    work page 2009

  31. [31]

    Graph embedded pose clustering for anomaly detection

    Amir Markovitz, Gilad Sharir, Itamar Friedman, Lihi Zelnik- Manor, and Shai Avidan. Graph embedded pose clustering for anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 10539–10547, 2020. 6

    work page 2020

  32. [32]

    Tree-based diffusion schr¨odinger bridge with applications to wasserstein barycenters

    Maxence Noble, Valentin De Bortoli, Arnaud Doucet, and Alain Durmus. Tree-based diffusion schr¨odinger bridge with applications to wasserstein barycenters. Advances in Neural Information Processing Systems, 36, 2024. 2

    work page 2024

  33. [33]

    Deep anomaly detection with deviation networks

    Guansong Pang, Chunhua Shen, and Anton Van Den Hen- gel. Deep anomaly detection with deviation networks. In Proceedings of the 25th ACM SIGKDD international confer- ence on knowledge discovery & data mining, pages 353–362,

    work page

  34. [34]

    arXiv preprint arXiv:2108.00462 , year=

    Guansong Pang, Choubo Ding, Chunhua Shen, and Anton van den Hengel. Explainable deep few-shot anomaly detection with deviation networks. arXiv preprint arXiv:2108.00462, 2021. 2, 5, 6

    work page arXiv 2021

  35. [35]

    Focal loss for dense ob- ject detection

    T-YLPG Ross and GKHP Doll ´ar. Focal loss for dense ob- ject detection. In proceedings of the IEEE conference on computer vision and pattern recognition, pages 2980–2988,

    work page

  36. [36]

    Multiresolution knowledge distillation for anomaly detection

    Mohammadreza Salehi, Niousha Sadjadi, Soroosh Baselizadeh, Mohammad H Rohban, and Hamid R Ra- biee. Multiresolution knowledge distillation for anomaly detection. In Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition , pages 14902–14912, 2021. 6, 1

    work page 2021

  37. [37]

    Diffusion schr ¨odinger bridge matching

    Yuyang Shi, Valentin De Bortoli, Andrew Campbell, and Ar- naud Doucet. Diffusion schr ¨odinger bridge matching. Ad- vances in Neural Information Processing Systems, 36, 2024. 2

    work page 2024

  38. [38]

    A public fabric database for defect detection methods and results

    Javier Silvestre-Blanes, Teresa Albero-Albero, Ignacio Mi- ralles, Rub ´en P ´erez-Llorens, and Jorge Moreno. A public fabric database for defect detection methods and results. Au- tex Research Journal, 19(4):363–374, 2019. 6, 1

    work page 2019

  39. [39]

    Csi: Novelty detection via contrastive learning on dis- tributionally shifted instances

    Jihoon Tack, Sangwoo Mo, Jongheon Jeong, and Jinwoo Shin. Csi: Novelty detection via contrastive learning on dis- tributionally shifted instances. Advances in neural informa- tion processing systems, 33:11839–11852, 2020. 6

    work page 2020

  40. [40]

    Weakly supervised learn- ing for industrial optical inspection

    Matthias Wieler and Tobias Hahn. Weakly supervised learn- ing for industrial optical inspection. In DAGM symposium in, page 11, 2007. 6, 1

    work page 2007

  41. [41]

    Explicit boundary guided semi-push- pull contrastive learning for supervised anomaly detection

    Xincheng Yao, Ruoqi Li, Jing Zhang, Jun Sun, and Chongyang Zhang. 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, pages 24490–24499, 2023. 1, 2

    work page 2023

  42. [42]

    Hierarchical gaussian mixture normal- izing flow modeling for unified anomaly detection

    Xincheng Yao, Ruoqi Li, Zefeng Qian, Lu Wang, and Chongyang Zhang. Hierarchical gaussian mixture normal- izing flow modeling for unified anomaly detection. arXiv preprint arXiv:2403.13349, 2024. 1

    work page arXiv 2024

  43. [43]

    Cutmix: Regu- larization strategy to train strong classifiers with localizable features

    Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. Cutmix: Regu- larization strategy to train strong classifiers with localizable features. In Proceedings of the IEEE/CVF international con- ference on computer vision, pages 6023–6032, 2019. 6

    work page 2019

  44. [44]

    Anomalyclip: Object-agnostic prompt learn- ing for zero-shot anomaly detection

    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. 1

    work page arXiv 2023

  45. [45]

    Anomaly heterogeneity learning for open-set supervised anomaly detection

    Jiawen Zhu, Choubo Ding, Yu Tian, and Guansong Pang. Anomaly heterogeneity learning for open-set supervised anomaly detection. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 17616–17626, 2024. 1, 2, 6, 7

    work page 2024

  46. [46]

    Towards open set video anomaly detection

    Yuansheng Zhu, Wentao Bao, and Qi Yu. Towards open set video anomaly detection. In European Conference on Com- puter Vision, pages 395–412. Springer, 2022. 1, 2 Distribution Prototype Diffusion Learning for Open-set Supervised Anomaly Detection Supplementary Material

    work page 2022

  47. [47]

    Dataset Statistics Extensive experiments are conducted on nine real-world anomaly detection (AD) datasets. Tab. 4 provides key statis- tics for all datasets used in this study. We follow the exact same settings as in previous open-set supervised anomaly detection (OSAD) studies. Specifically, for the MVTec AD dataset, we adhere to the original split, divi...

    work page 2021

  48. [48]

    5 presents a comprehensive comparison of the pro- posed DPDL method with state-of-the-art (SOTA) ap- proaches under general settings

    Full Results under General Setting Tab. 5 presents a comprehensive comparison of the pro- posed DPDL method with state-of-the-art (SOTA) ap- proaches under general settings. It reports performance metrics for each category within the MVTec AD dataset. Overall, the DPDL model consistently outperforms base- line methods across all application scenarios in b...

    work page

  49. [49]

    Detailed Class-level AUC Results under Hard Setting To evaluate the performance of the DPDL framework in detecting emerging anomaly classes, we conducted exper- iments under challenging settings and provided detailed results on six multi-subset datasets, including per-class anomaly performance, as shown in Tab. 6. Overall, the DPDL model achieved the high...

    work page

  50. [50]

    The Algorithm of DPDL Algorithm 1 Distribution Prototype Diffusion Learning 1: Input: Input X = {(xi, yi)}, C, ϵ, κ 2: for epoch = 1 to n do 3: Extract features F feature ← − X 4: Distribution of normal samples transformPMGP bridge ← − P (F) 5: Distribution Prototype Learning LDPL = Ln DPL + La DPL 6: Dispersion Feature Learning LDFL 7: Sample xi ∼ X , ec...

    work page

  51. [51]

    (13) and (14) We use Eqns

    Derivation of Eqns. (13) and (14) We use Eqns. (8) and (12) to derive Eqn. (13) as follows: π(ψ(xn i )|xn i ) = 1 ϖ(xn i ) exp( ⟨xn i , ψ(xn i )⟩ ϵ ) CX c=1 αcN (ψ(xn i ); µc, σc) = 1 ϖ(xn i ) CX c=1 αc(2π)−D/2|σc|−1/2 exp( ⟨xn i , ψ(xn i )⟩ ϵ ) exp(−1 2(ψ(xn i )))⊤σ−1 c (ψ(xn i ) − µc)) = 1 ϖ(xn i ) CX c=1 αc(2π)−D/2|σc|−1/2 exp( 1 2ϵ(2xn i ⊤ψ(xn i ) − ψ...

    work page