Something about semantic segmentation

I use the pre-trained model resnet50 in torchvision to do semantic segmentation
Should I change rgb value from[0,255] to [0,1]

yes and use mean/std normalization values as described here:

Thank you!
I transform the data from[0,255] to [0,1],but the performace doesn’t change.
Especially in the begin iters ,loss doesn’t change(I think maybe the loss will be lower than when I use [0,255])