1.Introduction

1.1 Anomaly Detection

• Few anomaly data
• Training only normal data

1.2 Representation-based

• feature extraction module

  -Resnet/Transfomer
  

• distribution estimation module

  -Parametric metric: DNN/CNN/NF
  
  -Non parametric metric: kNN/Mahalanobis distance
  

• 참조

1.3 Related Model

1.3.1 SPADE

• Extraction feature of all layer
• Using Memory Bank