1 Introduction

1.1 What is AD?

Untitled

Untitled

Fig 1, 2 of MVTec AD, cvpr19

1.2 Applications

1.2.1 industrial inspection

  1. 集成电路领域的陶瓷封装基板CPS3D-Seg, AAAI26

  2. 环状变压器线圈

    image.png

  3. finding defects of objects or materials on industrial production lines [5, 7, 10, 15, 36, 56, 62, 76] — SSPCAB, cvpr22

1.2.2 public security

detecting abnormal events such as traffic accidents, fights, explosions, etc. [12, 13, 17–19, 27, 28, 33, 39, 41, 47–50, 52, 67, 72, 73, 77, 78]. — SSPCAB, cvpr22

Untitled

Untitled

fig 6 & 8 of SSPCAB, cvpr22

1.3 The problem

  1. fundamental principle: modeling normality in the training data and assessing whether a test image aligns with this learned normality (常态/正常性).
    1. also pursued as a one-class classification (outlier detection) problem
  2. The nature of AD problem: unsupervised + the restricted data access (the |normal sample| for training is not often large enough, 60 — 200, even k-shot or zero-shot)
    1. “Since abnormal samples are available only at test time, supervised learning methods are not directly applicable to anomaly detection.” is it true? Yes, in most case.
  3. Learn normality via feature embedding or reconstruction based methods