DT/RT Channel mixer same input, different results

darktable

(Olaf Berthelsen) #1

I load the RGB color wheel in DT/RT and set the red channel to zero in the channel mixer. The top red triangle becomes black (0, 0, 0) in DT as expected but a sort of green (0,0% 38,0% 6,7%) in RT.
I include the color wheel and screen shots.


(Ingo Weyrich) #2

Set working profile to sRGB


(Olaf Berthelsen) #3

Thank you for your response but this seems really strange to me…

According to Rawpedia you should not change you working profile for normal use. What is normal use? And (0, 0, 0) should be pure black with any working profile, should it not?

How do you change the working profile to sRGB and is it possible to do in 5.4? What are the consequences for your future photo editing?


(Alberto) #4

hi,

yes. but (1,0,0) in srgb is not (1,0,0) in prophoto. so, you think you are starting from (1,0,0), but if working in prophoto you are not. just put a color picker in the red region and you will see


#5

I understand that a saturated red (1,0;0) in a small gamut colorspace is not a saturated red in a wide gamut colorspace giving non nul g and b.
That said, it is also paradoxical and surprising that setting the red to 0 in channel mixer you obtain some mixed green and blue. It seems to defeat the channel mixer objective.
Is the working profile the good space to apply this function? Shouldn’t be applied on output profile?


(Olaf Berthelsen) #6

Hi’
I’m following the excellent thread:

Please help me to understand "Channel Mixer"

because, well yes, in order to understand the channel mixer. When editing photos I often feel a need to finetune colors and of course you can do that by changing white balance, saturation, vibrance and so on. But when I try some basic tools i.e. channel mixer and others it seems difficult to control/understand what is going on. That’s why I’m studying this subject………

It’s correct that there are some small differences in the color picker results for the red triangle before applying the channel mixer. My RGB results are as follows:

DT 2.6: 254, 0, 0
RT 5.4: 99,6% 0,0% 0,0%

But I don’t understand why this results in the following when setting the red channel to 0, 0, 0 in the channel mixer:

DT 2.6: 0, 0, 0
RT 5.4: 0,0% 38,0% 6,7%

This a big difference when processing the same input.
I certainly agree that

because you should be able to predict and understand what is happening when using the tool. I would very much appreciate a clarification of this…….:o)!

As of now it seems to me that the DT results are what you would expect…….


(Alberto) #7

Here’s what’s happening:

The behaviour seems quite consistent to me. You zero out the R channel, so you are left with the G and B ones, which are nonzero in the current working profile.

Why? By definition, it’s the working profile the one you are expected to be working with… RT does allow you to change the working profile, so I see no problem here. The only issue that I see is that you can only set the working profile globally, for all the operations. That’s a restriction of RT. If you need more, there are other fine tools around here that can help you (e.g. GIMP or Photoflow).


(Olaf Berthelsen) #8

Hi’
Thank you for an interesting response.
Working/output profiles seems to be a kind of specialist subject somewhat obscure or remote to an average user such as myself. But it is indeed interesting when you explain how it works. Thank you for your effort.

Can I ask you the following question: Imagine your photo is the RGB color wheel. The red triangle is obviously of a clear red color. Now you want to use the channel mixer to turn the triangle into black. Not necessarily absolute black but a “deep” black color. How would you do this in a simple an intuitive way. No calculations and color pickers please…….:o)!


(Alberto) #9

You are asking the wrong person :slight_smile:
I think the channel mixer is an advanced tool, and not necessarily intuitive. In fact, I was arguing that it should be moved to the “advanced” tab already over one year ago…:


(Olaf Berthelsen) #10

OK fair enough………:o)!

The current Rawpedia text on the channel mixer consist of 6 lines of text. This is not overdoing it, is it? I think that improvements are needed.

So I hereby pass on the challenge to the Rawtherapee or Darktable community in general: A simple and intuitive explanation or examples on how to use the channel mixer.


(Mica) #11

Often times the best explanations comes from those who initially don’t understand a tool, but take the time to figure it out then explain it :slight_smile:


(Olaf Berthelsen) #12

A good point…………

But even so I think I will welcome other contributions by creating a separate topic. Something like a challenge to a simple guide to the channel mixer with a reference to this and the other channel mixer thread.


(Olaf Berthelsen) #13

Hi’
A question on profiles. Reading your answer and som information on the web I think I understand the following:

The input RPG-file is read into RT and converted into the working profile which in RT is the Prophoto RGB color space. This causes no loss of information since this color space is very big. All work in RT is performed in Prophoto in all modules (also lab Adjustments?). The end result is then transformed into the output profile which is sRGB. sRGB is a smaller color space so some information is lost in this process.
Is this correct?

DT apparently works in a smaller color space and som information is thus lost when processing the image and the output is different form the RT output.

According to the documentation most DTs modules works internally in the lab color space which is even bigger than the Prophoto color space according to wikipedia! The expression “most modules” sounds a little bit suspicious to me and implies that conversion is taking part internally in DT between modules causing loss of accuracy?
Is this possible?


(Mica) #14

darktable works in LAB, I believe.


(Ingo Weyrich) #15

No, only a part of the toolchain is performed in working profile. From a certain point on, most of the processing is performed in Lab.

Lab tool, ToneMapping, Vibrance, Lab colour correction grid, Local Contrast, impulse denoise, defringe, Sharpening, Microcontrast, Contrast by Detail levels, Wavelet, Softlight, Resize Method Lanczos, post resize sharpening are all done in Lab


(Alberto) #16

softlight is in RGB.

regarding the other question on accuracy, it is a can of worms which I’m not interested in opening (again) :slight_smile:


(Ingo Weyrich) #17

You’re right. I was fooled by ipf.softLight(labView); :wink:


(Olaf Berthelsen) #18

Ok, thank you to everybody………

The result of all this seems to be the following in the eyes of a average user trying to understand the “channel mixer” tool:

The channel mixer is indeed not a simple tool to understand or to use when the objective is to fine tune your photos.
As proven in this case you can open the same simple jpg image in RT and DT and you do not observe differences looking at your monitor. When you then perform the same simple channel mixer adjustment (turning off the red channel) you can get very different results.
It seems to be impossible or very tricky to explain which result is best/most correct.

I think that all can agree on the following (in theory): Pure red turns to black when the red channel is turned off.

I know now that the red triangle of the test image is not absolutely pure red, but very close. When studying colors and how to mix colors you will of course have to use the test material available to you. If anybody can provide better test material I would be grateful.

Until then the DT result makes most sense to me.


(Morgan Hardwood) #19

Change the working profile in RawTherapee to sRGB or Adobe RGB, then both RT and dt look the same.


#20

My color science is very basic.

What can be fooling users is that the notion of pure red depends on the colorspace you use.
Here the colors of initial image are defined in the sRGB colorspace. This is a small colorspace compared to what colors human being can see (the so called horseshoe). and one of the smallest colorspace.

In the CIE chromaticity diagram, the colorspace is represented by a triangle. Pure colors (primaries) are on the summits of triangle.
Small trangles represent small spaces and large triangle large spaces.

So what you call “pure red” is only pure in the small sRGB space.
A “pure red” of your image is no longer pure in a wider colorspace (prophoto). And the “pure green” contains also red.
When you set in that wide space R=0, you project the point on the opposite side of the triangle. It becomes a color that is outside the sRGB space (sRGB triangle) and cannot be represented in this small space without additionnal processing ( see rendering intent).
So what seems simple at first sight becomes rapidly overcomplex when digging in the color management science.
See for instance https://ninedegreesbelow.com/ for more accurate informations.

Perhaps DT stays in the input space for this operation?