MediaBrain-SJTU / RegAD

[ECCV2022 Oral] Registration based Few-Shot Anomaly Detection
MIT License
277 stars 40 forks source link

Question about your training strategy. #17

Closed FantasticGNU closed 2 years ago

FantasticGNU commented 2 years ago

In your code, I noticeed that you do a test on the test set on every epoch, and save the best epoch. Maybe before finishing your training procedure, the test set shouldn't be seen? It seems like that you decide the epoch number depend on the test set, I think it's not good.

chaoqinhuang commented 2 years ago

Thank you very much. To have a fair comparison with other baseline methods, we choose to follow their experimental setting and make sure the comparisons are fair.