Journal of the Korea Institute of Information and Communication Engineering (한국정보통신학회논문지)

The Korea Institute of Information and Commucation Engineering (한국정보통신학회)
권호 표지
DOI QR코드

DOI QR Code

gMLP-based Self-Supervised Learning Anomaly Detection using a Simple Synthetic Data Generation Method

단순한 합성데이터 생성 방식을 활용한 gMLP 기반 자기 지도 학습 이상탐지 기법

  • Ju-Hyo, Hwang (Department of Smart Manufacturing Engineering, Changwon National University) ;
  • Kyo-Hong, Jin (Department of Electronic Engineering, Changwon National University)
  • Received : 2022.11.02
  • Accepted : 2022.11.18
  • Published : 2023.01.31
Download PDF
⟨ Previous Next ⟩

Abstract

The existing self-supervised learning-based CutPaste generated synthetic data by cutting and attaching specific patches from normal images and then performed anomaly detection. However, this method has a problem in that there is a clear difference in the boundary of the patch. NSA for solving these problems have achieved higher anomaly detection performance by generating natural synthetic data through Poisson Blending. However, NSA has the disadvantage of having many hyperparameters that need to be adjusted for each class. In this paper, synthetic data similar to normal were generated by a simple method of making the size of the synthetic patch very small. At this time, since the patches are so locally synthesized, models that learn local features can easily overfit synthetic data. Therefore, we performed anomaly detection using gMLP, which learns global features, and even with simple synthesis methods, we were able to achieve higher performance than conventional self-supervised learning techniques.

기존의 자기지도 학습 기반의 CutPaste 기법은 정상 이미지에서 특정 패치를 자르고 붙이는 방법으로 합성 데이터를 생성한 뒤 이상탐지를 수행하였다. 그러나 이런 방식으로 생성된 합성데이터는 패치의 경계에 뚜렷한 차이가 나타나는 문제가 발생된다. 이러한 문제를 해결하기 위한 NSA 기법은 Poisson Blending을 통해 자연스러운 합성 데이터를 생성하여 더 높은 이상탐지 성능을 달성하였다. 그러나 NSA 기법은 클래스마다 조정해야하는 하이퍼 파라미터가 많은 단점을 가지고 있다. 본 논문에서는 합성 패치의 크기를 매우 작게 하는 단순한 방법으로 정상과 유사한 합성 데이터를 생성하였다. 이 때 패치가 매우 지역적으로 합성되기 때문에, 지역적인 특징을 학습하는 모델을 사용하면 합성 데이터에 쉽게 과적합 될 수 있다. 따라서 전역적인 특징을 학습하는 gMLP를 사용하여 이상탐지를 수행하였고, 단순한 합성 방법으로도 기존 자기 지도 학습 기법보다 더 높은 성능을 달성할 수 있었다.

Keywords

Acknowledgement

This work was supported by the Manufacturing Process Innovation Simulation Center Establishment Project funded by Ministry of the Trade, Industry & Energy(MOTIE, Korea).

References

  1. H. M. Schluter, J. Tan, B. Hou, and B. Kainz, "Natural Synthetic Anomalies for Self-Supervised Anomaly Detection and Localization." in Proceedings of Compute Vision-ECCV 2022, Tel Aviv, Israel, pp. 474-489, 2021. DOI: 10.1007/978-3-031-19821-2_27.
  2. C. -L. Li, K. Sohn, J. Yoon, and T. Pfister, "CutPaste: Self-Supervised Learning for Anomaly Detection and Localization," in Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Nashville: TN, USA, pp. 9659-9669, 2021. DOI: 10.1109/CVPR46437.2021.00954.
  3. J. Song, K. Kong, Y.-I. Park, S.-G. Kim, and S.-J. Kang, "AnoSeg: Anomaly Segmentation Network Using SelfSupervised Learning." arXiv:2110.03396, 2021.
  4. P. Perez, M. Gangnet, and A. Blake, "Poisson image editing," in Proceedings of ACM SIGGRAPH 2003 Papers on - SIGGRAPH '03, ACM Press, San Diego: CA, USA, pp. 313-318, 2003. DOI: 10.1145/1201775.882269.
  5. P. Bergmann, M. Fauser, D. Sattlegger, and C. Steger, "MVTec AD - A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection," in Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach: CA, USA, pp. 9584-9592, 2019. DOI: 10.1109/CVPR.2019.00982.
  6. H. Liu, Z. Dai, D. R. So, and Q. V. Le, "Pay Attention to MLPs," in Proceedings of NeurIPS 2021.
  7. O. Ronneberger, P. Fischer, and T. Brox, "U-Net: Convolutional Networks for Biomedical Image Segmentation," in Proceedings of Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015, Munich, Germany, pp. 234-241, 2015. DOI: 10.1007/978-3-319-24574-4_28.
  8. A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dhgha,ni, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby, "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale," in Proceedings of ICLR 2021, Virtual, 2021.
  9. I. Tolstikhin, N. Houlsby, A. Kolesnikove, L. Beyer, X. Zhai, T. Unterthiner, J. Yung, A. Steiner, D. Keysers, J. Uszkoreit, M. Lucic, and A. Dosovitskiy, "MLP-Mixer: An all-MLP Architecture for Vision," in Proceedings of NeurIPS2021, Virtual, 2021.