1 Like
I’m using all my best tricks on this one, but I’m afraid I still didn’t go far enough. dt 3.2.1
20200717_Garzweiler_3278.DNG.xmp (9.4 KB)
These machines must be huge! I have an hard time trying to understand the true scale. A man, causally walking under the big wheel, would help. 
My take with dt 3.4.
20200717_Garzweiler_3278.DNG.xmp (13.2 KB)
1 Like
It is huge…
Have a look at the image at the top right (full size). You can see human sized doors to get an idea exactly how big they are.
1 Like
OMG, this is humongous! I didn’t even spot the humans immediately. 
Been a while since I have used darktable or any photo software for that matter. Decided on a sort of vintage b&w look because that machine just asks for it. Speaking of dt has come a long way I am really impressed.
20200717_Garzweiler_3278.DNG.xmp (11.1 KB)
Let’s try this file for experimenting with Sigmoid. After opening I see some trouble with the file and had to adjust the black offset here which I’m not used to do very often. But OK.
My reference edit with Filmic v5 gives a usable result. Sigmoid has much less control possibilities but looks also reasonable, but in my opinion Filmic looks still more natural when increasing contrast, especially when rendering the overexposed sky and the excavator, whereas increasing contrast in Sigmoid soon leads to the impression of “a bit overdone”.
This is the Filmic version:
20200717_Garzweiler_3278.DNG.xmp (6,8 KB)
This is the Sigmoid verison:
20200717_Garzweiler_3278_sigmoid.DNG.xmp (6,7 KB)
Both algos are really hard to match. In this picture Filmic gives much more “clearity” in the shades and better rendition of the colors of the sandy material in the background.
Here is a snapshot comparison (filmic left vs. sigmoid right) at the “same” contrast factor (1.8)
Btw., I captured myself a photo of the older excavator 258 in 2013 with some people at the wheel, which has a diameter of 15m. This is much smaller than the diameter of 21,6m of excavator 288, but still impressive.
1 Like
Thanks for posting and good to see someone else testing sigmoid. Here are some crops out of your edit:
Lighter part of the image:
filmic
sigmoid
While I think the differences are sublte, filmic delivers a bit more contrast here.
Darker part of the image:
filmic
sigmoid
filmic crushes the shadows a bit, while sigmoid shows a bit more details. This leads to a clearer look in this part of the image in the filmic edit. Watching the rendering of the lamps, I think sigmoid does a better job avoiding halos.
Both images also show a slightly different colouring, visable in the shadow areas but also in the sand and coal. In my edit (#22) I remember that this was the point where I struggled the most. Is the sand more red ore more yellow ? How black is the coal really ?
Here is another try, sigmoid contrast = 2.25, skew = -0.35. It gives a bit more contrast to the excavator and a bit more definition in the sandy area (history in the jpg).
To get even closer to filmics result you could tweak sigmoid a bit more, especially the colour processing parameters. Personally I think both edits are close enough and would do any further editing with other modules such as tone equalizer, color calibration, color balance rgb etc.
Edit: typos
filmic crushes the shadows a bit, while sigmoid shows a bit more details. This leads to a clearer look in this part of the image in the filmic edit.
That is because I set black-relative exposure in Filmic to only -2.09 (calculated by color picker) which is probably a little bit to high, although I’m still able to distinguish the shades of black at my Eizo CS2731. A value in the range [-3.00 … -2.50] gives a bit more headroom in the dark blacks, but with a value below -3.00 the output picture starts lacking global contrast. At the end, its just about personal preferences.
Filmic black -2.09 (left) vs. -3.00 (right)
Watching the rendering of the lamps, I think sigmoid does a better job avoiding halos.
Those lamps have clipped highlights and I’m using Filmics reconstruction, but it doesn’t help here. Therefore Filmic produces a reconstructed area around a sharp circle of blown highlights, based on the settings. Unfortunately I’m not able to find a suitable setting here. Sigmoid, which hasn’t reconstruction, just clips and may look a bit better here. In both cases I find the magenta CA at the lamps on the left machine more anoying than the blown highlights.
To get even closer to filmics result you could tweak sigmoid a bit more, especially the colour processing parameters.
Yes, you’re right. In vector scope (Jz) you clearly see a color shift from blue towards red when changing the color processing in Sigmoid from crosstalk to rgb ratio + rgb power norm, which finally compensates the blue color cast.
1 Like
In my experience, sigmoid has a smoother gradient in very light areas, e.g. like in this case around areas with sensor clipping. This avoids the typical halos or artefacts of filmicrgb. With the reconstruct tab in filmicrgb you can correct this to a certain degree but this needs a bit of experience and extra care.
See:
Taken from: https://share.streamlit.io/jandren/tone-curve-explorer
Maybe @jandren can answer this with a more theoretical background.
1 Like
These curves visualize the mapping for non sensor-clipped pixels. Filmic let you define the upper (and lower) bound of the curve, by white (black) relative exposure slider, and various other parameters, and its possible to create asymmetric curves. Sigmoid let you define something like „the center point“ of a symmetric curve by the skew slider, if I understand that correctly…
Sensor-clipping is a different beast to handle, as the composition of the light spectrum was lost during capture and needs to be reconstructed from valid information of other channels and/or surrounding pixels.
@pass712 and @rkowalke I will chip in with what I know so far.
Highlights! I think the reason the highlights look weird in filmic and not in the sigmoid has its roots in the color processing mode. Changing from crosstalk to preserve hue will result in similarly weird transitions in the spotlights as for the filmic edit. Meaning that changing the preserve colors to “no” in filmic should help with this particular problem in filmic. I think the proper solution for this problem is to upgrade the highlight reconstruction module to something better than the current solution which was developed in the pre-scene-referred era of darktable. That said, the sigmoid does not clip, it never does, it converges towards peak luminance as the pixel luminance goes towards infinity.
Filmic does in contrast clip at the black and white point. So adding a black point of -2.09 means a very very aggressive clip of the blacks. The contrast setting is not numerically comparable between sigmoid and filmic so just do what pleases your eye with that setting.
More about the shadows, both crushness in filmic and weird color in sigmoid. I think this is caused by the heavy usage of black level compensation in the exposure module. My recommendation is to try other options before fiddling with that setting as it can have some odd effects on the image. The original image has very low contrast and there is evidently a lot of fog present. Just look at it without sigmoid/filmic!
I could push the contrast quite far in sigmoid when I tried to edit the image myself. Just added a bit of decreased saturation of the sand in the background. Also went for the slightly negative skew as that gave a bit more interesting shadows and kind of made it a bit more off a dusk feeling.
I feel like the mist is still very present in this image. I feel like adding a bit of local contrast made it slightly better at least. But be careful with the local contrast as it changes the color slightly and doesn’t do so well in very bright regions as it operates on only the L channel of Lab and is developed for the display referred workflow.
1 Like
.
