New OCIO colour management

[Joelaff]

OK. That has been my approach too over the last 30+ years. Always fun explaining gamuts to clients.

Regarding color illusions I used to get into that with a client, where he would start telling me that part of version A looked different from version B (parts that were not changed). Didn't take me long to realize the issue.

I said, "You have version B on the left and version A on the right of your screen don't you?"

"Uh, yeah, how did you know?"

"Swap them..." I said.

(Now the effect was reversed because it was all perceptual based on the colors around the area he was examining. Blows people away when they realize how relative color is.)

[piotrus3333]

n't we want to use the inverse of whatever output transformation we intend to use in the end?

if footage is colour graded it would be even harder.
lets say it's RED footage. starts in camera, RED wide gamut something something. goes to RED log3G10, colour grading, and then through standard RED log to 709 lut, which by itself is different look than ACES look.
it might not be noticeable with textures used for illumination but if the image is a plate with cg on top, when you reverse transform this plate to ACEScg assuming it started as ACES SDR video - the plate might just not match to cg part - black levels might be off for example.

not something that is about to ruin your tomorrow but worth keeping in mind.

[Joelaff]

So let me see if I understand. Are you saying that there is some footage that if we were to run it through the ACES SDR Video RGC Colorspace to ACEScg and then apply the ACES SDR Video RGC again would not match the input? Even if it started out in the Rec709 gamut (or sRGB for that matter)?

I thought that anything in Rec709 or sRGB to begin with was a subset (sub-gamut) up anything ACEScg, and without wild transforms losing precision in 32bit float (in theory possible) would come back out the same.

In my quick test of a test image saved out as a PNG in Rec709 and brought back into Nuke set to the Rec709 ACES SDR Video RGC Colorspace (with Nuke set to ACEScg colorspace as scene linear) then saved back out through the rec709 ACES SDR Video RGC I can find not even one digit of difference in Nuke's readout.

[piotrus3333]

it will match the input - as you know now ACES transforms are easy to invert. this was apparently one of design choices in the beginning.

imagine a car paint, dark saturated red. you have a perfect, scanned material in VRay, acescg textures and all that.
you are rendering the car on a plate with more such cars. if the plate is in camera log - it is perfectly invertible to linear colour in one of camera specific wide gamut spaces - data from the plate match data from your ACEScg render 1:1 (assuming obviously ideal scenario: reds are identical)
if this plate goes through even gentle grading and a display transform that is different enough from ACES display transform - you end up with much narrower gamut so different colours, also different contrast (ACES display transform is much too dark in the shadows for many) etc.
now when you render your perfect, scanned VRay mtl of said car paint on that plate, that you inverted with ACES display transform - you might find out that the reds are not the same - your scanned red is still as good as it was but scene referred colour data obtained by reversing display referred rec709 plate with transform (or chain of transforms as data from camera or renderer travel a long way before it is moulded to be ready for screen) other than it was created with no longer match "reality".

pixels on screen are only tiny slice of scene referred data and it would be safer to always assume it is never invertible than otherwise. log footage (or aces exrs) = less cgi work.

[Joelaff]

Yeah, I pretty much assume nothing ever matches reality. No two lights match, and no light is perfectly white balanced. Filaments and even LEDs age. LEDs aren’t true black body, white cards / color checkers fade, etc., etc.

Hell, human color perception is influenced by numerous physical factors, caffeine, etc. and this is in constant flux.

Coming from the dark ages of film I don’t expect exact matches, I just need to get things in the close ballpark to then be colored artistically by myself or others.

I appreciate all the effort you have put into learning about ACES and sharing with others. Thanks.