Validation and Testing in Anomalib

[Samyarrahimi]

I am trying to understand the validation and testing phases in Anomalib. Assuming we are using MinMax Normalization and F1AdaptiveThreshold, In Anomalib, during validation, the model tracks the minimum and maximum anomaly scores across samples to be used for MinMax normalization. Additionally, it calculates an optimal threshold using the F1AdaptiveThreshold to classify samples.

During testing, the model uses these min and max values from validation to normalize predicted anomaly scores, and it classifies each sample using the threshold determined during validation.

How does the model classify samples during validation when the min and max values for normalization and the F1AdaptiveThreshold are still unknown? How does it assign prediction labels to validation samples?

is the model iterating through the validation set twice? once for collecting the minimum and maximum scores, and once for normalizing the scores, finding the optimal threshold, and classifying the validation samples for metric calculations(validation precision, recall, ...)?

I would appreciate answers that include references to the source code, if possible.