Method to normalize the image between -1 to 1. Is it mean-std norm of imagenet or anything else?

[ashishkumar822]

Hi.

In the paper it is mentioned that image is normalized between -1 to 1. However I could not find any line in the script which does this.
Although I saw Dlimp augmentation function but I could not find -1 to normalization.

My question is, which kind of normalization has been used:

1. subtract with imagenet mean and divide by imagenet std
2. Min-max norm to bring the image in 0 to 1 and then img = 2 * (img - 0.5)
3. Or simple img = (img - 127) / 128 if image is in 0 to 255.

These details are import for finetuning. Please provide a calarity on this. Thank you.