Helping understanding paper FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows

Hello Family.

I am working on this paper FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows

I have some question that will post here.
The first question is about what two paragraphs say

For ResNet, we directly use the features of the last layer in the first three blocks, and
put these features into three corresponding FastFlow model.


For ResNet18 and Wide-ResNet50-2, we directly use the features of the last layer in the first three blocks, put these features into the 2D flow model to obtain their respective anomaly de-
tection and localization results, and finally take the average value as the final result.

Focusing on the feature level that i need to extract, do I need to extract one or three features levels?
I understood to extract one feature which implementation looks like this:

self.feature_extractor = models.wide_resnet50_2(True)
self.feature_extractor = create_feature_extractor(
self.feature_extractor,
{‘layer1’: ‘Feat’}
)

Hello Miguel, did you solve it? I am working on this paper too

I am still working on it.

Are needed 3 NF and layers [1, 2, 3] from resnet. Cheers

1 Like

Do you know which is the loss function? I can’t identify it in the paper

Thank you.

Hello Andres. I highly recommend checking this discussion