search

dhkdnduq / PatchCore_anomaly_detection

Unofficial implementation of PatchCore anomaly detection
Apache License 2.0
25 stars 4 forks source link

Accuracy #4

Closed royarahimzadeh closed 3 years ago

royarahimzadeh commented 3 years ago

Hi, I ran your code with some categories, but the auc roc results are not the same as the ones reported https://github.com/hcw-00/PatchCore_anomaly_detection For example, 'toothbrush' category should be 100% accuracy, but I get Area under curve Precision Recall: 0.96 Would you please tell me what should be the problem? I also have another question about the weighted score. Did you consider that reweighting in your code?

Thank you

dhkdnduq commented 3 years ago

1. Main:Calculate by pixel per pixel using gt. This project : Use the average value of the image.

For industries that cannot use gt.

2.If there's more data, feature.pt gets bigger too. It's doesn' mean reweight.But I think I do it instead.

royarahimzadeh commented 3 years ago

Thank you.