Editing moments with darktable

[anon41087856]

@s7habo

Then I used Filmic RGB to adjust the dynamic range so that the details in the snow can still

On that step, it is quite obvious that your histogram still has some space on the right, so you could have been more aggressive on the white exposure. Also, since you later darken the snow to an actual middle grey-ish, you could have directly raised the middle grey in filmic to spare you some steps later.

@s7habo

Now comes the tricky task of emphasizing the details without affecting the dynamic range too much. Here you suggest to use a local contrast module. Although this will actually help to increase local contrast a bit it is not enough and there is a risk that you will either over-sharpen the image or get other unwanted effects. Note that the snow has not yet gained much in local contrast:

Local contrast is good at emphasizing details around mid-tones. So again, putting the snow closer to middle-grey in filmic should help you there.

@s7habo

Now we use the RGB curve to increase the details in the snow. I found this to be a difficult task, as it works differently than the normal curves in darktable. Even very small movements of the points can cause big changes in the image and it is not easy to get a reasonable contrast in the medium grey area .

The contrast in color balance allows you to define a fulcrum very easily, so you can set whatever grey shade will be the reference for your contrast enhancement without having to worry about an off-balanced UI.

@s7habo

So, as far as details in the snow are concerned, we are almost there. I have found that in certain situations Haze removal module is very suitable for highlighting the details and is even much better than local contrast module.

Again, local contrast would have been easier if the snow had been closer to middle grey. Haze removal is really grounded into physics (water droplets diffusion), it can backfire real quick if used outside its base assumptions.

@s7habo

But you can’t simply use a module at the beginning of the pixelpipe again, because these changes would have affected the results of all other modules that come later.

That’s not true for all modules. Actually, the beauty of the scene-referred workflow is that any module doing linear operations on linear RGB can be reordered freely around other linear modules and still produce the exact same result (unless you use parametric masking).

@s7habo

If I increased the contrast with color balance module after step 2 in the example above and want to use it again after step 6, it would not be recommended because the contrast focus was moved in between .

Yeah, but that’s easily corrected by adjusting the contrast fulcrum accordingly.

@s7habo

The same applies to tone equalizer, for example. The mask for guided filter, which was made at the beginning, is no longer valid after a few steps in the pixelpipe .

That’s actually wrong. The mask of the tone equalizer is computed very early in the pipe, right after exposure module. Whatever you do later in the pipe has no effect on it.

@s7habo

It is still unclear which modules can (should) be combined with each other to avoid artefacts. As far as I have noticed, exposure, filmic RGB tone equalizer and rgb curve belong to the linear workflow .

See the answer I gave here: 3.0 How to get good results automatically? - #4 by anon41087856

@s7habo

I found it much easier to use the tone mapping module and limit its effect to the darkest areas of the image than to have to deal with the not so easy masking in tone equalizer again.

The problem of tone mapping is it does a bilateral blur on log data, which is bound to produce halos. It’s just a bad algorithm.