search

StevenLiuWen / ano_pred_cvpr2018

Official implementation of Paper Future Frame Prediction for Anomaly Detection -- A New Baseline, CVPR 2018
430 stars 126 forks source link

low AUROC but good AUPR #32

Open ethanyhzhang opened 5 years ago

ethanyhzhang commented 5 years ago

As I retrained the model on UCSD and avenue datasets, the result of AU of ROC on UCSD looks fine, but the AU of ROC on avenue is only 0.67 but AU of PR on avenue reached 0.8493. Having made sure the evaluate_name is correctly. BTW, the psnr value is about 10 for avenue, but about 37 using your provided pretrained model. Wonder if there is somewhere different for the setting to train on avenue? Thx.

a726 commented 4 years ago

As I retrained the model on UCSD and avenue datasets, the result of AU of ROC on UCSD looks fine, but the AU of ROC on avenue is only 0.67 but AU of PR on avenue reached 0.8493. Having made sure the evaluate_name is correctly. BTW, the psnr value is about 10 for avenue, but about 37 using your provided pretrained model. Wonder if there is somewhere different for the setting to train on avenue? Thx.

Hello, I avenue in the training data set, the psnr is 10, you know that this is caused, how to solve? How you should lose in the training data set shanghaotech command? Thanks.