byungjae89 / SPADE-pytorch

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

Different results on "transistor" #6

Open TzxzT opened 3 years ago

TzxzT commented 3 years ago

Hi, byungjae: I just run your code, interestingly, I find a same problem mentioned before: the result of "transistor" is different with yours, while others are same. I get 0.910 auc vs 0.903 auc. (https://github.com/byungjae89/MahalanobisAD-pytorch/issues/2#issuecomment-757613795) In that issue, the version of efficientent_pytorch is different. However in current code, WRN50-2 in pytorch zoo is used. Maybe the model in pytorch zoo also changed before, but I can't understand why "transistor" is quite different?