Normalisation of G-Buffers

[WouterBesse]

Hello, thank you for this fantastic work!

I was wondering if you are normalising your gbuffers in any way.
I'm running into an issue where for example my position gbuffer has a mean of 11.2 and sd of 30.7, which I think might be messing up my result.

I thought of just normalising each gbuffer individually, but since this is absolute data one position value could mean a completely different position than in another normalised position gbuffer.

Would the solution be getting the standard deviation and mean of my whole dataset, and normalising the group, or am I seeing problems where they aren't actually there?

Thank you :)

[WouterBesse]

Currently I am just individually normalising all the gbuffers, independently from eachother. This seems to improve the training of the model greatly. However, I will mark this as closed once I also tested the group normalisation to see what the difference is.