search

piergiaj / pytorch-i3d

Apache License 2.0
971 stars 250 forks source link

Loss function #35

Open XingyuJinTI opened 5 years ago

XingyuJinTI commented 5 years ago

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?

piergiaj commented 5 years ago

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

XingyuJinTI commented 5 years ago

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?