Normalization using pre-trained vgg16 networks

I am trying to use vgg16 network for my own task dataset. Since vgg16 is trained on ImageNet, for image normalization, I see a lot of people just use the mean and std statistics calculated for ImageNet for their own dataset.

I am not sure which is more reasonable. What is your experience of using mean and std computed on your dataset as opposed to using ImageNet mean and std?