the output of I3D model and the calculation of loss function

[punknownq]

Hi,
Thank you for your work, firstly.
Do we need to add x = x.mean(2) in forward function of the I3D model?
And why do you add if config.model == 'i3d': targets = torch.unsqueeze(targets, -1) in train_epoch while it has not been added in validation_epoch?