search

nelson1425 / EfficientAD

Unofficial implementation of EfficientAD https://arxiv.org/abs/2303.14535
https://arxiv.org/abs/2303.14535
Apache License 2.0
240 stars 62 forks source link

Question about the loss between student output and ae output #11

Closed Youskrpig closed 1 year ago

Youskrpig commented 1 year ago

Hi， I'm puzzled about the loss between student output and ae output，is it necessary?

nelson1425 commented 1 year ago

Hi, yes it is necessary in my experience. In the paper the author says "The student learns the systematic reconstruction errors of the autoencoder on normal images, e.g., blurry reconstructions. At the same time, it does not learn the reconstruction errors for anomalies because these are not part of the training set. This makes the difference between the autoencoder’s output and the student’s output well-suited for computing the anomaly map."