Question about unbounded processing

Thanks for the inputs here, in particular from @AD4K and @flannelhead, which confirms some muddled thoughts of mine – I’m now confused on a steadily higher level.

One question regarding the difference between scene- and display-referred: I’ve understood an important aspect of this to be that not only is the representation of luminance differences linear in scene-referred, but also that it is not bounded in computations. Since nobody mentions this, I wonder if one considers bounded representation to be inherent in non-linear representation – or have I misunderstood here?

From my understanding the data in DT are processed in 32bit float so never bounded but if you use a module with a display referred UI then you can only modify data in that bounded 0-1 range… At least I believe someone told me this or I have messed it up… :slight_smile:

I believe I read something to that effect in here…

I moved this over to a new thread, since it diverges from the original topic but deserves more attention even though it has been discussed before.

The difference between scene-referred and display referred is in what the numbers mean.
In “scene-referred”, pixel values are proportional to light energy,
in “display-referred”, those values are proportional to log(light energy), which has (nasty) effects when you use blurs (as in most masking).

I don’t think there’s any reason why display-referred should be bound, that may just be an historical artifact: as a traditional display can’t receive values greater than 100%, the modules concerned weren’t really concerned with values outside the range (0…1). In programming, giving a display-referred module an input >1 would lead to “undefined behaviour” (i.e. anything can happen).

It is not. Jzazbz color space / Jacob Rus | Observable is, for example, nonlinear and unbounded

I learn something new. This is interesting and I will keep it in mind when masking.

Is not a blur on the display’s linear light is good?

I really have no idea what your question is…

1 Like

Glad I am not the only one…This is border line bot like… looking back over the few posts by this user they are mostly weird one line questions… sorry @slow if you are real…

I am real. I mean it works good blurring on display linear light values? It is good to me?

1 Like

This can be tested objectively in newer versions of GIMP or other applications. I am sure there were several posts on this very question, but unfortunately, I cannot find it anywhere!

I find this article by Elle Stone to be most definitive: