Input normalization in new unified CV models

[bhack]

With the emerging unified CV models we need to start to think to another type of preprocessing as the transformation required to "normalize" different inputs in a common representation (and eventually "denormalize" the outputs):
https://unified-io.allenai.org/
https://arxiv.org/abs/2206.07669

Are we going to plan something in the Keras-cv roadmap?

Please add any relevant resource here.