search

byungjae89 / SPADE-pytorch

PyTorch implementation of "Sub-Image Anomaly Detection with Deep Pyramid Correspondences"
Apache License 2.0
235 stars 43 forks source link

The threshold for labeling image to normal or anamoly #7

Open letmejoin opened 3 years ago

letmejoin commented 3 years ago

As mentioned in paper, "Positive classification is determined by verifying if the kNN distance is larger than a threshold t", but when I check codes, I couldn't find the threshold. At the test time, you compute all the test image otherwise the anamoly images. So, if I missunderstand?

momo1476422482 commented 2 years ago

Personaly, I found that the aht author of the repo used the ground truth mask of test images to get the threshold, which is ....wired...

Avocadoooooo commented 1 year ago

Personaly, I found that the aht author of the repo used the ground truth mask of test images to get the threshold, which is ....wired...

I have the same doubt