Scene-referred and L*aburves are aded

Now how to do that in scene-referred without changing color too much if at all.

You shot that pretty dark for an indoor shot… I wonder if you could just start with a brighter image and then less correction is needed… also if you use a color checker and use that for wb and color you might not have to worry so much about how your exposure and tone are changing color… I guess its comes down to what your perception of what the shot should look like and how to best represent that ???

I used the Sekonic chart middle gray to create an accurate reference under my 5000K lighting. L*=49.7 a=0.03 b=-2.2 So, the color is correct. I could definitely start with a brighter image by increasing exposure +1ev. I’m not so sure that it would make any difference though after calibrating the chart.

I had to increase the exposure 1.418 ev to calibrate the chart. If I ettr the shot 1.33 ev and re-calibrate, would that help to hold color when doing scene referred adjustments?

Me neither. One of my Gurus, Elle Stone, said:

ā€œScene-referred RGB data is bounded only by Chromaticity.ā€ … in other words, Lightness is not.

Way to technical for me. Anyway thanks for sharing.

The operative word in that article is ā€œboundedā€, she shows that values of Luminance Y due to editing can not be clipped if someone goes over the top.

Whereas display-referred for 8-bit monitor automatically clips anything over 255.

Maybe L* in CIELAB does not correlate with Y in CIEXYZ.

I have no idea what darktable means by ā€œscene-referredā€, so my comment was a general one.

I was using a display-referred module (tone curve) within a scene-referred workflow. Would it be okay continue doing that as long as brightening was kept well under 255 which is always the case for my hobby photography?

Sorry, Mike, I don’t use darktable and have no idea what each ā€œmoduleā€ does or how it does it. In principle, never exceeding 255 is OK unless there are specular highlights (e.g. sunlit motorbike) that you don’t care about. I’m less sure about ā€œkept well under 255ā€, not knowing how much ā€œwell underā€ is.

1 Like

Yes. Its fine. It isn’t 100% optimal, but it works. If you want the best results, make sure Tone Curve comes after Filmic/Sigmoid.

1 Like

@paperdigits Advise taken. I’m just surprised that scene-referred doesn’t have a module that preserves color when lightening or darkening. At least that’s my impression.

Its a complicated topics, and I’m no expert, but generally when you change the value (lightness or darkness) you also change the perceived color as well. I’m sure there are plenty of people who can explain it better than I can.

I’m not a technical person so I’m sure that further explanations would most likely be over my head.

This might be a moot point, but having a L*ab adjust module catered to scene-referred might be useful for certain photos where maintaining color accuracy is important.

I don’t think that is possible.

DT uses 32-bit float and tried for the most part to keep operations linear and with a fairly close proportional relationship to the data in the scene… Then it uses a tone mapper to manage that and bring it to the display… So you have all that light…you add or subtract exposure to give you middle gray ie put your scene middle grey at that point in the display DNR… Then filmic or sigmoid will map the data on either side of that as you specify…

THere is a dynamic display of this in filmic

image

A potential filmic-based rendering. Left: your sidecar, with the tone curve turned off (ground truth for colours). Right: filmic with color balance rgb.

But you are correct, the dynamic range of the image is so small that there is no need to compress it with a tone mapper; it’s all reflective surfaces. Your tone curve already shows that: the histogram does not come near black and it stays far away from white. In fact, your curve expands the dynamic range (it increases contrast almost everywhere, with the shoulder being above the original input, only a bit of the toe affecting the darker portions in the original.
sRGB histogram without your curve:
image
With your curve:
image

Alternatively, your version (with the tone curve) vs sigmoid with the smooth preset:

But again, both with filmic and sigmoid, here we are just making a contrast adjustment.

(Note that the screenshots are not completely what I see on my display, as they are in my monitor’s colour space, which does not match sRGB completely, and I did not convert them from that space to sRGB manually. They are not that far off.)

This would be a good place to start:

It also explains the issues with Lab.

The problem is that you have no real control over what happens to values above 1. Anything that changes the shape of the right side of the curve could have unintended consequences. Have you tried Tone EQ with the simple tone curve preset? It’s supposed to work exactly like a tone curve, but of course compatible with unbounded data.

And this part of the manual would probably be useful reading: darktable 4.6 user manual - darktable's color dimensions

Thanks, Todd!

Interesting that it also seems to involve display-referred actions, so not as described by the ICC folks:

The above links to:

https://www.color.org/scene_analysis_and_rendering.xalter

I just looked at that. I don’t see any change when I select that preset. Am I missing something?

Capture

darktable 4.8

No guided filter… see in masking for preservation of details…