out of the box RAW rendering not up to JPEGs

MStraeten · January 18, 2020, 11:51am

this chart wasn’t used to create the profile … but with the applied style (lut+tonecurves) it matches the ooc jpeg…

asn · January 18, 2020, 12:27pm

The tone curve is just created by comparing the grey ramp and setting the points accordingly in the tone curve. It has nothing to do with the color reproduction …

It will match JPGs which are made under similar lightning conditions.

anon41087856 · January 18, 2020, 12:38pm

Both are intimately correlated. The tone curve alone will affect the saturation in non-linear ways. Then, the LUT will revert that.

asn · January 18, 2020, 12:47pm

@anon41087856 Isn’t the tone curve calculated and then unapplied on the color patches?

anon41087856 · January 18, 2020, 12:48pm

I don’t think so. That would make no sense.

asn · January 18, 2020, 12:55pm

github.com

darktable-org/darktable/blob/master/src/chart/main.c#L1012-L1049

    
      
          #else // rgb tonecurve affecting colours, too
            tonecurve_t rgbcurve;
            // ownership transferred to tonecurve object, so we just alloc without free here:
            cx = malloc(sizeof(double)*num_tonecurve);
            cy = malloc(sizeof(double)*num_tonecurve);
            cx[0] = cy[0] = 0.0;                           // fix black
            cx[num_tonecurve - 1] = cy[num_tonecurve - 1] = 100.0; // fix white
            for(int k = 1; k < num_tonecurve-1; k++)
            {
              float DT_ALIGNED_PIXEL rgb[4], Lab[4] = { 0.0f, 0.0f, 0.0f };
              Lab[0] = grays[6*k+0];
              dt_Lab_to_prophotorgb(Lab, rgb);
              cx[k] = rgb[0];
              Lab[0] = tonecurve_apply(&tonecurve, Lab[0]);
              dt_Lab_to_prophotorgb(Lab, rgb);
              cy[k] = rgb[0];
            }
            tonecurve_create(&rgbcurve, cx, cy, num_tonecurve);
            free(grays);

This file has been truncated. show original

anon41087856 · January 18, 2020, 1:08pm

Inverting means applying the inverse curve, and not every curve is mathematically invertible (it needs to be a bijection). Unless some part of the curve building ensures strict monotonicity, this code is dangerous.

Nor_man · January 19, 2020, 8:13pm

I like to share my workflow with DT3.0:

I created different profiles for my camera to match the ooc jpg rendering. (Muted, Neutral, Portrait Landscape…) This way I can apply different styles to one picture and can choose which looks best for further processing. You even can combine those styles with filmic rgb if you choose settings like this (which are nearly complete neutral):

then you can compress dynamic range with white relative exposure or lighten up midtones with middle grey luminance.
@anon41087856 it would be nice if we could set black relative exposure to -18 or maybe -20 or do you have other settings which have really zero effect to the image?

For the profiling with darktable chart i set the input color profile to linear rec709 RGB because this way I later have absolutely no trouble with those blue LED Lights (like the bridge). But I need two or more instances of color look up table (always with 49 patches) to get results nearly perfectly close to ooc jpg. I run darktable-chart a second or third time with the C-LUTs created before. (I recommend to correct the first C-LUT manually if needed in DT)

The finished style looks like this:

And this is one of the results (left jpg - right RAW):

for me its the best compromise:

different good looking starting points really close to ooc rendering in one step
no messing around with “ugly” blue colors
can use filmic and toneequalizer which are huge improvements
If included those user created styles and auto-applyed to matching RAWs: great first experience for dummies using DT the first time

edit: here are the files if someone wants to have a look:
https://drive.google.com/file/d/1PKPaAawwNaxJTLZTuun3MbFn2pOGBCik/view?usp=sharing
https://drive.google.com/file/d/10Yr3jEiGdJReI6cZJB7euwuyETzcW_PF/view?usp=sharing
https://drive.google.com/file/d/1LqUkjwIYkBG7KIXSwZsLKunx_NpvSety/view?usp=sharing
CC BY-SA 4.0

anon41087856 · January 19, 2020, 11:54pm

That tells me you didn’t understand the purpose of the tool. No current camera has more than 14.9 EV dynamic range at 64 ISO, so these values don’t make sense. Logarithms can’t deal with zero input (what is zero energy ? a black hole ?), and filmic shouldn’t be pushed that far. For zero effect on the image, disable the module.

Nor_man · January 20, 2020, 8:26am

sorry, I have expressed myself incomprehensibly. what I do is combine the color lookuptables and tonecurve created with darktable-chart with filmicRGB. the only task in this workflow for filmic is to rescale the dynamic range without adding contrast.
for this I need a preset in filmicRGB which initially has almost no influence on the picture. I have deactivated the curve in the “look” tab of filmic as far as possible. in the “scene” tab the settings as shown in the screenshot above are the most neutral (but not 100%) I have found. from this point I can now rescale the dynamic range as required.

i know tonecurve is working in Lab and this is sometimes problematic but it works 95% of all cases. if it doesn’t work I still have the option to deactivate the tonecurve and set everything with FilmicRGB.

Entropy512 · January 20, 2020, 2:21pm

I really dislike this, as in many cases, it causes people to blame the camera for deficiencies as opposed to “undocumented Adobe magic”. For example, Sony cameras, when using LR or C1 default profiles, are widely regarded as having issues with rendering of non-Caucasian skin colors. In general, it’s something that goes away if you profile the camera yourself.

anon41087856 · January 21, 2020, 12:26am

Set contrast to 1 in filmic look tab, then latitude to 100%, then save that as a preset.

priort · January 26, 2020, 3:35am

@anon41087856 I have seen in your video that you recommend LC after Filmic but I see it actually comes before in the pixel pipe. Is that because it was too hard to change without messing again with the module order…if it gets dragged after and you save that as a style will it still execute after or will the style use the default order?? Just wondering…

Matt_Maguire · January 26, 2020, 3:54am

LC should indeed come after Filmic RGB in the pixel pipe, and this is what I observe in the default pixel pipe order in DT 3.0.0. Which version are you using @priort?

Peter · January 26, 2020, 7:04am

Local contrast is in my version of darktable above RGB Filmic (and therefore after in pixel pipe).

elstoc · January 26, 2020, 8:01am

If you imported the image in an earlier version of darktable that image will retain the old pixelpipe order (so as not to change the effect of previous edits you might have made). To get the new order you currently need to discard the history stack.

If you want to do this and retain previous edits then duplicate the file, discard history in the duplicate, then copy the history stack from the original to the duplicate.

There will hopefully be a new module soon that allows you to update the history stack order to match that of a specific version of darktable.

asn · January 26, 2020, 4:08pm

The camera profile you created with darktable-chart is normally for daylight images. That’s when you created the profile.

For daylight images you took, just apply the LUT and click in filmic rgb on auto tune level and you’re done.

priort · January 30, 2020, 5:20pm

@Matt_Maguire @Peter @anon41087856 Fresh edit and now on 3.1+420…it is as it should be …so maybe I was looking at an old edit or had an older version at the time…all good LC now seems to be after filmic for me…

peleme · February 13, 2020, 9:04pm

@Nor_man
Thank you so much for your work! The Oly_PM_Natural dtstyle helped me a lot with my em5 mark iii
(Vielen Dank für Deine Mühe und das Du die Ergebnisse zur Verfügung stellst)

priort · March 31, 2021, 3:14pm

This is an old post I stumbled on …may I ask for clarification…as you use 49 patches you have a chart with more than 24 I would guess…For your multiple CLUT do you run chart once. Then generate an new jpg from the raw with only the CLUT of the first run or do you apply the tone curve. Then you run this again once or twice…so then you use the final tone curve and keep the 3 luts created from each iteration??