I feel trying my method of comparing the camera’s JPG to the RAW file edit would be a reasonable comparision. Neutral ICC I would expect a very dull and lifeless picture because that would be my description of the neutral profile for the cameras JPG. I would probably use standard as my target ICC profile.
At what stage in the pipeline is it best to make the comparison. I am not looking to match the JPG, just trying it out to see if it is in any measureable way “better”, meaning a better representation of the actual colours that are easier to process I suppose.
I am finding not after these initial tests, darktable already does a very good job with the standard input profile.
I guess I’m not doing a good job of explaining my perspective…
The “actual” colors of a capture come from a color transform that’s anchored to a measure of how people perceive color, that is, what “color” they assert a particular mix of light to be. That for most cameras is anchored in the CIE 1931 color-matching experiments, where 17 people were presented with various spectral mixes and were tasked with matching them with a gizmo that allowed them to mix (roughly) R, G, and B lights to do the match. Every mainstream consumer camera I know uses this reference to anchor their R, G, and B dyes on the mosaic.
So, the baseline for renders that “accurately” present colors that represent what folk saw at the scene would be a profile that transforms the large camera response to equivalent hues that the rendition media is capable of presenting. If you want to know that you’re getting that, you would use a transform that was trained on color patches measured to the CIE 1931 reference. Nikon doesn’t tell you that about any of their profiles, so best you can do is compare their renders to ones you know are produced to the CIE 1931 reference.
I think you find dt doing a good job of it because the internal data it uses is largely sourced from target-shot-CIE-measured profiles. Gotta be a little skeptical, though, sometimes software developers will insert a set of numbers from Adobe profiles, some of which are good, others not so much. Based on my cursory inspections, I don’t think dt has any like that, but RT does.
This is the important part: If you want to improve on what the camera captured, you need to start with that 1) CIE-referenced color, 2) scene-linear tone rendition. We haven’t talked yet in this thread about #2, that gets into the filmic-sigmoid-whatevertonecurve discussion. But, if you start with anything else, a big part of your work will probably be trying to undo what was added, rather than improving…
I am comparing the jpg from the word go. For instance DT’s exposure module makes a presumption of raising exposure by 0.7 for every camera. But the actual amount may need to be different. For my G16 it is 0.0 and my R7 is about +1.0EV correction needed. I also find having color calibration module set to as shot in camera brings the color in close but then certain color zones may need lifting in either color zones or color equalizer module. With my R7 I found the need to lift the chroma of the warmer tones near the red zones to better match the JPG look. Also the shadow and highlights and local contrast module defaults help the result.
I must stress that I am not trying to simply chase the JPG look believing it is my ultimate goal, but this approach has given me a better starting point which is often good enough for a one click edit. I believe you are just trying to also chase a better starting point for your images. If a ICC profile gets that for you I would be very keen to hear about that approach. I just haven’t personally gone down that rabbit hole to see if it will work well. But I am interested in how this works for you and if it works I may also try that out as I have not yet taken the opportunity to profile my Nikon cameras.
My X-rite mini color-checker came with a leaflet listing the sRGB, the CIELAB and the Munsell values for each patch, stating the lighting for each list.
In the past, I shot the color-checker under cheap LED strip lighting, of course knowing that to be incorrect.
Next, I opened the resulting capture in RawTherapee and white-balanced on the mid-gray patch and and adjusted the brightness until the L* for the mid-gray patch was about 50 then recorded the Lab* values for the patches of interest, usually BGRYMC (3rd row).
Then transferred the results to a spreadsheet already loaded with the leaflet values and which calculates the a*b* delta-E for each patch of interest.
Good for comparisons rather that absolute accuracy.
1 Like
I never liked that crude approximation, so my computer remains Adobe-free and I normally use the proprietary raw converter on the basis that they know better than I …
1 Like
This script works pretty well actually. I have in the past made a couple of profiles for nikon and sony users and they seemed to like them…I don’t use it routinely…but it might actually offer a better starting point than trying to adopt those nikon profiles…
Most of us are using D65 camera profiles when we let dt/RT/Art/whatever use their internal table of camera primaries, and things look just fine no matter the lighting. We take care of that with the white balance multipliers instead.
I say the use case for anything like that would be color reproduction work, images of paintings, etc. For that, I’d want to train a profile with measured patches from the subject, a lot more work than is desirable for most anything else.
Digital imaging is a whole chain of “good 'nufs”, we need to remember that as we peeps our pixels…
Edit: Oh, for the color repro use case, a decent compromise would be to make a profile using a target shot illuminated by the same light as the scene.
1 Like
I had an hour at the local nature reserve this afternoon with my Tamron Adaptall-2 90mm SP, had it a few months and this is the first day out with it.
I tried the Nikon ICC profile again, and still I prefer the DT standard profile. Here is the photo (using the DT standard profile) my XMP and the Nikon Neutral ICC if anyone wants to try it out.
DSC_5102.nef (18.6 MB)
DSC_5102_01.nef.xmp (21.4 KB)
Nkn_D300_neutral.icm (235.2 KB)
Out of interest, I tried the Nikon profile on one of my images in the GIMP, using assign i.e. without changing the image data:
Image at right has the GIMP standard sRGB.
Nikon at left; I am not impressed - Easter Red Cedar trunks are not that color.
<opinion deleted>
I think they’re input profiles, not working profiles?
1 Like
Sorry, can’t help; I don’t really know the difference between those and embedded ICC profiles or each other for that matter.
Input “profiles” describe the meaning of the R, G and B values, just like any other colour space. But input profiles are for raw files only (and spedific to brand/model). Once you have treated the raw files, the images are transformed into a standard colour space (sRGB, Rec2020, …).
And since the images are not encoded in those raw ICC colour spaces, assigning one in the GIMP will always give strange results, So will assigning e.g. Rec2020 to an image encoded in sRGB…
1 Like
Yes, the Nikon profile is an input profile.
[edit] I’m beginning to realize how little I know about profiling in general, so I’m butting out of this thread. Please ignore my previous ramblings …[/edit]
Just realised that I didn’t upload the actual NEF file, which was taken with a D300. I will upload it to my post with the image in tonight.
Do you think there will be any interest in uploading it as a playraw, along with the Nikon ICC?
Playraw is good, but we don’t want to host the nikon ICC as you probably violate something by distributing it.
1 Like
Understood, didn’t realise that! I have uploaded the NEF to my post with the image in it.
I have come to the conclusion that using the extracted Nikon ICC profiles from Nikon Studio (Capture NX or whatever) is not worth the trouble. I do wonder if it has something to do with the fact that the unbreak input colour module works in display referred space is an issue, and that it doesn’t work with the scene referred workflow. No idea, out of my depth at this stage.
It’s not like I can’t get excellent results as is, just wanted to try it out again.
1 Like
Presumably only talking about darktable?
RawTherapee can open many types of file and has options including <none>, <embedded> or <custom> for ‘input profile’ in the Color Management module for any type of file.
No. RawTherapee can also use the same input profiles. I assume many others can to. Usually RT users would prefer a DCP rather than an ICC input profiles, I assume.