You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have been trying to reproduce the RegNet results, but unable to go past 91 on CIFAR, if you still have access to it would you mind sharing the learning rate you used?
Also, I noticed that for RegNetY_400MF, you use the same configuration as RegNetX_400MF with se_ratio=0.25, however, the official implementations uses a different configuration albeit the design was based on performance on imagenet -- was this change from the paper intentional based on performance on CIFAR?
Thanks for maintaining the repo.
The text was updated successfully, but these errors were encountered:
Hi ,
I have been trying to reproduce the RegNet results, but unable to go past 91 on CIFAR, if you still have access to it would you mind sharing the learning rate you used?
Also, I noticed that for RegNetY_400MF, you use the same configuration as RegNetX_400MF with se_ratio=0.25, however, the official implementations uses a different configuration albeit the design was based on performance on imagenet -- was this change from the paper intentional based on performance on CIFAR?
Thanks for maintaining the repo.
The text was updated successfully, but these errors were encountered: