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
@ildoonet
Thank you for a good suggestion.
Similar to the answer for #3 (comment) , we consider the efficiency for loading data and also simplicity of implementation, so we fix the lambda and the position of the random region for all the images in the mini-batch.
We didn't check the performance improvement when we change the lambda and cropping position, we will be very happy if you train CutMix with that mechanism.
After verifying, please PR with configure option (e.g. '--diverse-lambda' or '--diverse-position').
https://github.com/clovaai/CutMix-PyTorch/blob/master/train.py#L233
Can it be improved if we use different(diverse) lambdas for each data?
The text was updated successfully, but these errors were encountered: