Skip to main content

​
Introduction

EfficientAD introduces a fast patch descriptor and trains a student network to predict the features computed by a pretrained teacher network on nominal training images. Because the student is not trained on anomalous images, it generally fails to mimic the teacher on these. A large distance between the outputs of the teacher and the student thus enables the detection of anomalies at test time.

​
Key Features

  • Speed: Designed for fast inference, making it suitable for real-time applications
  • Teacher-Student Architecture: Uses knowledge distillation to learn normal patterns
  • Patch-based Detection: Operates on image patches for localized anomaly detection

​
How It Works

  1. Teacher Network: A pretrained network that extracts features from images
  2. Student Network: Trained to mimic the teacher’s output on nominal (normal) images
  3. Anomaly Detection: When the student fails to match the teacher’s predictions, an anomaly is detected
The distance between teacher and student outputs serves as the anomaly score - larger distances indicate higher likelihood of anomaly.

​
Advantages for Production

EfficientAD is well-suited for deployment in environments with limited memory and storage. Due to its relatively decreased reliance on storing nominal samples compared to methods like PatchCore, it behaves better in cold start scenarios where limited training data is available.
Edit this page on GitHub or file an issue.