I’ve been using Darktable for a while now and love the power and flexibility it offers.
However, with a recent photo taken with my Fuji X-T2, I noticed that the colours seemed a bit off compared to the SOOC jpeg in one particular area. On the top-right of the image near the edge, there is a band of cloud that is more orange than red, when I import this image into DT or RT, this comes out much more red than orange for some reason (scroll down to the DT screenshot to see the area I’m talking about).
RawTherapee renders the colours in the same way as Darktable.
I have also just downloaded the free version of Capture One, just to see what it does with the image and it renders the colours similar to how the camera renders them.
Here is the SOOC jpeg:
Here is my processed DT image:
And here is a screenshot from within DT with the history stack rewound to the very start before any modifications, showing the problem area:
And for comparison, here is a rough edit with Capture One Express:
As an aside, I’m not overly happy with my DT edit on this image, I had a battle with highlight recovery and disabled the base curve to try to get some detail back. Although I like the image, don’t judge me too much on my editing in this case!! I usually get better results with DT.
Were you using a particular film emulation for the SOOC jpeg?
If the default base curve doesn’t work for you, either adjust it by hand or switch to a different preset, like the Leica preset, which renders everything a bit darker.
As you have the edit now, you can probably use the Color Zones module to tweak the upper right corner to match the SOOC jpeg.
what settings are you using for RT? this is what I get when using the “Auto-Matched Curve - ISO Low” profile, seems pretty close to the camera JPG to me.
First of all, you can’t talk about color accuracy, as the SOOC jpeg is not “color accurate”, the one you showed has Velvia film simulation, and maybe did you also tweak the saturation and contrast in camera on top of that.
Anyway, here’s a version which I think quite close to the SOOC jpeg:
Rawtherapee: XT2S5663.jpg.out.pp3 (10.5 KB)
I own an X-T2 as well, and in RT I have a dynamic profile which loads an X-T2 dcp profile, but deactivates the built-in Tone curve and Look table, and instead loads a custom Tone curve in perceptual mode in the Exposure tab.
That’s basically what I used to process your raw file, but since your SOOC jpeg had Velvia applied, I also applied a Velvia Film simulation (HaldCLUT from this excellent set: https://discuss.pixls.us/t/new-fujifilm-xtrans-film-simulations/6333) and profiled lens correction.
Hi Mica, I almost never disable the Base Curve but I have found that the RAW overexposure warnings show that you can pull back some blown highlights on occasion, and this was one of them.
I would usually bracket a shot like this and then use Hugin to create a tif to bring back into DT but on this occasion, I hadn’t bracketed.
I’m not very familiar with RawTherapee although I do have it installed. I usually use Darktable.
The setup you have with your X-T2 sounds very interesting Sebastian, I’ll have to look into that and see if something similar can be done in DT. I got some film simulation presets from a post on here a few months ago, they’re very good.
The redness of that part might be because of your custom base curve. It works in RGB color space that means it can cause hue shift (more information here and here). If you share your sidecard (the XMP file) for your edit, it would help to investigate the reason more precisely.
Euan, I showed what I get in RT because you mentioned RT in the thread’s title. I’m not trying to convince you to use RT, I know that it’s difficult to switch to another software.
When you have a very intense red-orange, the colors might be in a ratio of 4:1:0 red:green:blue.
However, when it’s very bright, the red channel alone gets clipped and the resulting color ratios can become 4:2:0, much more orange, or 4:4:0, which is fully yellow.
This is objectively incorrect; sunsets are far more red in real life than commonly depicted in photographs.
So in this case, the Fuji and Capture One results are unrealistic, and darktable is much more faithful to the original scene.
There are many reasons why a JPEG image produced by a camera (or by manufacturer software) may differ from that produced by other software. Remember that a raw file is like a kitchen full of raw ingredients, while a JPEG is baked by a specific chef.
RawTherapee 5.4 uses a color matrix derived from dcraw for converting sensor values into the XYZ color space for the FUJIFILM X-T2. The JPEG image embedded in the raw file might use a different matrix (I haven’t checked).
A camera performs various adjustments to the image when creating the JPEG. These adjustments are aesthetic in nature - accuracy is not of primary concern. These adjustments can also contribute to differences between the out-of-camera/embedded JPEG and what you see in other software.
The camera may be working in a different color space than other software.
In addition to the rigid requirements imposed by various standards concerning developing a raw photo there are various gray areas where there is no one correct way of handling things. Such areas may be handled in one program (your camera’s firmware) in one way and in another program (RT/dt) in another way, and neither is incorrect. For example, how are ultra-saturated colors to be handled - clip, desaturate, or shift hue? How is the input profile’s LUT to behave when the required adjustments are too acute - is it to literally reproduce known points at the risk of sharp transitions, or should it be relaxed to ensure smooth transitions at the cost of altering the values of known points, and if so, how should those value be altered? Different solutions will lead to different results.
Indeed. I totally forgot that I’m actually using the X-T20 dco profile bundled in RT, which I renamed to X-T2. Both caleras use the same sensor and processor and should behave similarly with regards to colors.
oh, right! I’m using one that I built myself from dpreview’s studio shots, as I usually do when there’s no better one in RT… it takes 5 minutes and the results are typically quite reasonable
