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

Screw Results Replication #4

Open bernhard2202 opened 3 years ago

bernhard2202 commented 3 years ago

Hello,

Thanks for publishing this code! I am currently trying to replicate the results from your paper with this repository. However, I struggle a lot with the screw-subset which is with the code from here around 0.7 auc and in the paper reported as 0.96 auc. Could you please give me any advice on how to close this gap?

Cheers, Bernhard

byungjae89 commented 3 years ago

Hello, @bernhard2202

There are two different auc exists, image-level auc and pixel-level auc. Would you check which type of auc you consider? screw's image-level auc is 66.7 and pixel-level auc is 99.1 in this repo.