I’m working with 3D convolutional network and I have to train ( from scratch) the net on a subset of UCF101. Do you know what are the normalization values ( mean, std) that I should to use with UCF101? Can I use the normalization values of ImageNet for UCF101? Thanks
The ImagNet stats might be a good starter, as the UCF101 dataset uses images from the same “domain” (natural images) and these stats are also used in tutorials such as this one.
You could of course also compute the stats from the training set manually and compare them to the ImageNet ones.
Thanks you for the reply !