Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
110 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

SoftPatch: Unsupervised Anomaly Detection with Noisy Data (2403.14233v1)

Published 21 Mar 2024 in cs.CV, cs.AI, and cs.LG

Abstract: Although mainstream unsupervised anomaly detection (AD) algorithms perform well in academic datasets, their performance is limited in practical application due to the ideal experimental setting of clean training data. Training with noisy data is an inevitable problem in real-world anomaly detection but is seldom discussed. This paper considers label-level noise in image sensory anomaly detection for the first time. To solve this problem, we proposed a memory-based unsupervised AD method, SoftPatch, which efficiently denoises the data at the patch level. Noise discriminators are utilized to generate outlier scores for patch-level noise elimination before coreset construction. The scores are then stored in the memory bank to soften the anomaly detection boundary. Compared with existing methods, SoftPatch maintains a strong modeling ability of normal data and alleviates the overconfidence problem in coreset. Comprehensive experiments in various noise scenes demonstrate that SoftPatch outperforms the state-of-the-art AD methods on the MVTecAD and BTAD benchmarks and is comparable to those methods under the setting without noise.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (44)
  1. Generalized out-of-distribution detection: A survey. arXiv preprint arXiv:2110.11334, 2021.
  2. Deep learning for unsupervised anomaly localization in industrial images: A survey. IEEE Transactions on Instrumentation and Measurement, 2022.
  3. Memseg: A semi-supervised method for image surface defect detection using differences and commonalities. arXiv preprint arXiv:2205.00908, 2022a.
  4. Dsr - a dual subspace re-projection network for surface anomaly detection. CoRR, abs/2208.01521, 2022.
  5. 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, 2022.
  6. Self-supervised predictive convolutional attentive block for anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13576–13586, 2022.
  7. Learning semantic context from normal samples for unsupervised anomaly detection. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 3110–3118, 2021.
  8. Towards total recall in industrial anomaly detection. arXiv preprint arXiv:2106.08265, 2021.
  9. Cflow-ad: Real-time unsupervised anomaly detection with localization via conditional normalizing flows. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 98–107, 2022.
  10. Focus your distribution: Coarse-to-fine non-contrastive learning for anomaly detection and localization. In 2022 IEEE International Conference on Multimedia and Expo (ICME), pages 1–6. IEEE, 2022.
  11. 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.
  12. Vt-adl: A vision transformer network for image anomaly detection and localization. In 2021 IEEE 30th International Symposium on Industrial Electronics (ISIE), pages 01–06. IEEE, 2021.
  13. A hierarchical transformation-discriminating generative model for few shot anomaly detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 8495–8504, October 2021.
  14. Cutpaste: Self-supervised learning for anomaly detection and localization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9664–9674, 2021.
  15. Draem-a discriminatively trained reconstruction embedding for surface anomaly detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8330–8339, 2021.
  16. Uninformed students: Student-teacher anomaly detection with discriminative latent embeddings. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4183–4192, 2020.
  17. Multiresolution knowledge distillation for anomaly detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14902–14912, 2021.
  18. Anomaly detection via reverse distillation from one-class embedding. arXiv preprint arXiv:2201.10703, 2022.
  19. Encoding structure-texture relation with p-net for anomaly detection in retinal images. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XX 16, pages 360–377. Springer, 2020.
  20. Omni-frequency channel-selection representations for unsupervised anomaly detection. arXiv preprint arXiv:2203.00259, 2022.
  21. Padim: a patch distribution modeling framework for anomaly detection and localization. In International Conference on Pattern Recognition, pages 475–489. Springer, 2021.
  22. Patch svdd: Patch-level svdd for anomaly detection and segmentation. In Proceedings of the Asian Conference on Computer Vision, 2020.
  23. Same same but differnet: Semi-supervised defect detection with normalizing flows. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1907–1916, 2021.
  24. Panda: Adapting pretrained features for anomaly detection and segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2806–2814, 2021.
  25. Cfa: Coupled-hypersphere-based feature adaptation for target-oriented anomaly localization. arXiv preprint arXiv:2206.04325, 2022.
  26. Divide-and-assemble: Learning block-wise memory for unsupervised anomaly detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8791–8800, 2021.
  27. Simple: Similar pseudo label exploitation for semi-supervised classification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15099–15108, June 2021.
  28. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. Advances in Neural Information Processing Systems, 33:596–608, 2020.
  29. Dividemix: Learning with noisy labels as semi-supervised learning. arXiv preprint arXiv:2002.07394, 2020.
  30. Resolving training biases via influence-based data relabeling. In International Conference on Learning Representations, 2021.
  31. End-to-end semi-supervised object detection with soft teacher. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3060–3069, 2021.
  32. Unbiased teacher for semi-supervised object detection. arXiv preprint arXiv:2102.09480, 2021a.
  33. Class-aware contrastive semi-supervised learning. arXiv preprint arXiv:2203.02261, 2022b.
  34. Adbench: Anomaly detection benchmark. arXiv preprint arXiv:2206.09426, 2022.
  35. Self-trained deep ordinal regression for end-to-end video anomaly detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12173–12182, 2020.
  36. Rca: A deep collaborative autoencoder approach for anomaly detection. In IJCAI, pages 1505–1511, 2021b.
  37. Anomaly detection with robust deep autoencoders. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, pages 665–674, 2017.
  38. Understanding and mitigating data contamination in deep anomaly detection: A kernel-based approach. In IJCAI, pages 2319–2325, 2022.
  39. Self-supervise, refine, repeat: Improving unsupervised anomaly detection. 2021.
  40. Latent outlier exposure for anomaly detection with contaminated data. arXiv preprint arXiv:2202.08088, 2022.
  41. Data refinement for fully unsupervised visual inspection using pre-trained networks. arXiv preprint arXiv:2202.12759, 2022.
  42. Deep one-class classification via interpolated gaussian descriptor. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 383–392, 2022.
  43. Leif E Peterson. K-nearest neighbor. Scholarpedia, 4(2):1883, 2009.
  44. Lof: identifying density-based local outliers. In Proceedings of the 2000 ACM SIGMOD international conference on Management of data, pages 93–104, 2000.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (9)
  1. Xi Jiang (53 papers)
  2. Ying Chen (333 papers)
  3. Qiang Nie (25 papers)
  4. Yong Liu (721 papers)
  5. Jianlin Liu (9 papers)
  6. Bin-Bin Gao (35 papers)
  7. Jun Liu (606 papers)
  8. Chengjie Wang (178 papers)
  9. Feng Zheng (117 papers)
Citations (45)
View on Semantic Scholar

Summary

Unsupervised Anomaly Detection with SoftPatch: Tackling Noisy Data Challenges

Introduction

Anomaly detection (AD) is a critical challenge in various domains, especially in industrial applications where detecting abnormal patterns is key to ensuring product quality. Traditional unsupervised AD algorithms have shown impressive results in academic datasets; however, their performance drastically decreases when faced with real-world noisy data. This discrepancy is mainly due to the unrealistic assumption of clean training datasets in most academic settings. To bridge this gap, we introduce SoftPatch, a novel unsupervised AD method designed specifically to address the challenge of noisy data in image sensory anomaly detection.

The Core Challenge with Noisy Data in AD

The fundamental difficulty with noisy data in anomaly detection lies in the algorithms' overreliance on the assumed cleanliness of the training dataset. In practical settings, especially in industrial environments, it is almost inevitable that the training data will contain some level of noise—be it due to inherent data variation or human error in labeling. This noise significantly affects the algorithms' ability to accurately model normal data behavior, leading to unreliable anomaly detection.

SoftPatch: A Novel Approach to Unsupervised AD

SoftPatch addresses the issue of noisy data by introducing a memory-based, unsupervised AD method that operates on a patch level to efficiently denoise the data. This approach contrasts sharply with traditional methods that work on a sample level, thereby improving the utilisation of data. The primary contributions of SoftPatch can be summarized as follows:

  • Introduction of Patch-Level Denoising: SoftPatch deploys noise discriminators to score and eliminate noise at the patch level, rather than at the image level. This granular approach allows for the retention of normal patches within noisy samples, enhancing the model's ability to accurately represent the normal dataset distribution.
  • Memory Bank with Soft Anomaly Detection Boundary: The model uses a memory bank to store the anomaly scores of patches, which softens the anomaly detection boundary by re-weighting coreset samples based on their outlier scores. This strategy effectively counteracts the model's overconfidence issue, a common pitfall in existing AD methods.

Experimental Validation and Results

Our experiments on the MVTecAD and BTAD benchmarks demonstrate SoftPatch's superior performance in various noise settings. Notably, SoftPatch not only outperforms state-of-the-art AD methods in noisy scenarios but also maintains comparable performance in clean settings. This adaptability showcases SoftPatch's robustness and its potential to cater to more practical, real-world AD applications.

Implications and Future Directions

The success of SoftPatch in tackling noisy data within unsupervised anomaly detection opens up several avenues for future research:

  • Exploration into Other Domains: While the current paper focuses on image sensory anomaly detection, the principles behind SoftPatch can potentially be adapted for use in other domains where noisy data is prevalent.
  • Integration with Semi-Supervised Techniques: Investigating how SoftPatch could be used in conjunction with semi-supervised learning techniques might yield even more robust anomaly detection models, capable of leveraging limited labeled data to further refine the detection process.
  • Scalability and Efficiency Improvements: Future work could also look into optimizing SoftPatch's computational efficiency, making it more scalable for handling larger datasets commonly found in industrial applications.

Conclusion

SoftPatch represents a significant step forward in unsupervised anomaly detection, particularly in the context of noisy data. By addressing the limitations of existing methods and introducing a novel patch-level denoising strategy, SoftPatch not only enhances the robustness of AD models but also extends their applicability to more realistic, noisy settings. As the field of anomaly detection continues to evolve, it is anticipated that SoftPatch and its underlying principles will inspire further innovations, driving the development of more reliable and generalized AD solutions.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets