pytorch-i3d icon indicating copy to clipboard operation
pytorch-i3d copied to clipboard

Published 20 hours ago • verified publisher icon piergiaj

Loss function

Open XingyuJinTI opened this issue 5 years ago • 2 comments

Hi there,

Thanks for the fancy work on I3D.

Just wondering, the loss function in the code is joint classification and localization loss. However, in the I3D paper is only classification loss, isn't it. Is there any reason why you change it?

XingyuJinTI avatar May 30 '19 14:05 XingyuJinTI

I changed it for training on Charades according to the authors method for charades.

piergiaj avatar May 30 '19 17:05 piergiaj

Do you mean the paper for Charades dataset. Thanks.

By the way, just to make sure that to compute classification loss you pick the max for matrix 'labels' across time, which is actually the same as directly using the class ground truth annotations, right?

XingyuJinTI avatar May 30 '19 18:05 XingyuJinTI