search

sungwool / CFA_for_anomaly_localization

99 stars 20 forks source link

Low performance compared to the Table 4 of the paper. #7

Closed JinyongShin closed 2 years ago

JinyongShin commented 2 years ago

When I run your code, I see that the performance is lower than that listed in Table 4 of the paper. In particular, in the case of the Screw class, the image AUROC is 97.3% in the table, but I got a value of 90.6. What could make this difference? Also, how can I get a similar value to the paper?

sungwool commented 2 years ago

In some classes, performance depends heavily on random seed. I haven't considered the impact of this depending on the experimental environment.

You can bypass the problem by changing the random seed of trainer_cfa.py.

Thank you.

JinyongShin commented 2 years ago

By setting args.Rd = false, and random seed = 777, I got a value of 96.6(%) for the image AUROC in the case of the screw class. Thank you.