Output regression standardisation

Hi! I’ve just started coding with PyTorch and I’ve built a regression model of a ResNet that, given an image 1x128x128, outputs a tensor made by 3 values (tensor([1,1,3]). The network should therefore predict the 3 values of the tensor. Being two of them always way larger than the other (e.g. [512, 520, 100]) I was thinking to apply a standardisation to the targets. How can this be done in PyTorch?
Hoping it’s clear what I mean. Thank you so much!