Camera gamut outside horseshoe?


(Morgan Hardwood) #21

You will find the opposite conclusion in the DCamProf documentation:


(Elle Stone) #22

Hmm, well, maybe 24 colorchecker patches is enough for a LUT profile if, as the dcamprof documentation says, “DCamProf will automatically relax the LUT a bit, that is prioritize smoothness over accuracy.”

Not having a colorchecker, there’s not much I can do to test the idea of making a LUT profile using only 24 patches. But it seems like an odd/iffy thing to do, even though I’m sure the author of DCamProf has carefully considered all aspects of making a camera input profile.

Experimenting with IT8 target charts, I had to relax the LUT quite a bit to get xiccly “-fif -g” curves that looked sufficiently smooth to me, and that didn’t also create color artifacts when applied to test images.

Perhaps the pigments in a colorchecker-24 chart really are so much superior to an IT8 chart that they make good LUT profiles even with only 24 patches. That still doesn’t mean a LUT input profile is better for everyday use compared to a linear gamma matrix profile for a camera that is assumed or tested to be linear in its reponse to light.

If anyone actually has both an IT8 and a colorchecker-24, it would be interesting to make a matrix and also a Lut profile using each and compare resulting colors to the other.


(Morgan Hardwood) #23

Plenty of shots on imaging-resource.com


(Elle Stone) #24

My apologies, I looked on that website more than once and never figured out what to download, actually not that long ago. I have a Sony A7 “mark I”. If you or someone could provide links to the specific files to download, I’d be happy to do some testing and report back. I don’t mean a single link to the whole heap of A7 files. I mean links to download the specific one or two files that contain the two target charts.

But I really don’t expect anyone to do this sort of “paw through the files” for me! I’m just saying I already gave that more effort than it was worth to me personally.


(Morgan Hardwood) #25

Usually you search for a review of the camera model of interest, then click on the “Samples” tab, and then on the “Thumbnails” link under the tab bar to find links to raw photos of color targets. Not always but sometimes you will find shots of more than one target type.

SONY ILCE-7:


(Elle Stone) #26

There are approximately 23 rows of thumbnails that look like there’s one or two target charts, plus another 8 or so rows of different thumbnails that look like possible candidates. I found that page before, but which raw file(s) should be downloaded? I’m assuming ISO 100, but which ev and which f-stop?

This is the point where I was stymied the last time I looked at that page. Plus where does one get the actual reference file?


(Graeme W. Gill) #27

Yep. Which is why it really only becomes a good approach using camera spectral sensitivities, where you can synthesis as many samples covering all the surface color gamut, with the statistics that you want. 2 D or 2.5 D cLUTs are also probably desirable, since exposure in cameras is an independent variable. (i.e. you make a 2D lookup with the 3rd dimension being either linear or at least orthogonal to the transformations in the chomatic dimensions.)

Something on my list of “things I’d like to try sometime” is using a chart measurement (perhaps taken under several different types of lighting) to estimate spectral sensitivities (based partly on “typical” camera curve shapes), and then use that to synthesize virtual samples to setup a 2.5 D cLUT.


(Graeme W. Gill) #28

Hmm. Less overlap between red and green tends to indicate being even less similar to the human observer, and so likely to have greater disparity with what we see.

But on reading the article about the Trichromatic back, I think you have misinterpreted what it is - it seems from the claims that in fact it is closer to the standard observer, and therefore will have more overlap than a normal back. This could potentially lead to more red-green noise, but should be better in overall color rendition.

[ But this article seems to indicate not a whole lot of difference after profiling. So maybe not that much of a difference really ? ]

That doesn’t follow either, if there is in fact more overlap not less. One of the commenters speculates that it is instead due to less sensitivity to UV & IR, which would follow for a sensor that more closely mimics human spectral sensitivities. Another reason would be the camera channels having less stray sensitivity in the other bands.


(Elle Stone) #29

I’ve been searching through the ArgyllCMS mailing list archives, and so far the only statement about making a profile using the ColorChecker 24 is that 24 patches isn’t a lot of patches:

But that thread was back in 2008. @gwgill - has a consensus emerged over the years that the ColorChecker 24 is suitable for making a LUT camera profile?


(Glenn Butcher) #30

I would be interested in helping with that, although it might be like the help one gets from one’s children when doing yard work…


(Graeme W. Gill) #31

I’m afraid time and motivation are the obstacles. I’m deep into a challenging new project, and it will take a while to wrap up (at least 12 months to go I’d say). And unfortunately one of the hard lessons I’ve had to learn is that giving software away for free is a terrible way of making a living :frowning:


(Graeme W. Gill) #32

Depends on the type of theater lighting though. If it’s “rock and roll” lighting using LED or moving light dichroic luminaires, then yes :-). Less garish lighting design or older style luminaires using gels or gel scrollers should be less of a problem, although the latter are really going out of favor in modern theater productions.


(Graeme W. Gill) #33

I wouldn’t say so, no. Good for matrix profiles, and easier to ensure other factors are good (i.e. uniform lighting, low flair etc.), but AFAIK most successful cLUT input profiles have used custom charts with many more patches or spectral sensitivities.


(Glenn Butcher) #34

Sadly yes, very much so: https://www.informationweek.com/it-life/ntps-fate-hinges-on-father-time/d/d-id/1319432. Hmmm, I started to write a bit on this, but that would be a thread hijack; I’ll start a new one. (Edit: did some research, my grand idea turned out not so grand. Still cogitating it, though…)


(Graeme W. Gill) #35

Yes, that story certainly hit home to me.


(Glenn Butcher) #36

Here’s my blue story, apologies if already posted.

First, with dcraw processing:


Yuck.

Next, with the Adobe Standard DCP for my camera, converted to an ICC with dcamprof:


Better, but still yuck.

Now, with my sunlight_calibrated camera profile, modified per the discussion at: https://www.ludd.ltu.se/~torger/dcamprof.html#make_profile_deep_blue


Note the blue has desaturated a bit, but it no longer ‘glops’. Also, note the letters, they’re now greener.

I didn’t use the dcamprof -y parameter; instead, I converted my profile from ICC to dcamprof JSON, opened the JSON file in an editor and changed the blue Y value by hand, then turned the JSON back into ICC with dcamprof makeicc. This is not where I want to end up, so there’s more experimenting in the works…


(Morgan Hardwood) #37

Best exposed lowest ISO, unless of course you’re testing the effects of noise on a profile. Same rules apply as when shooting your own target.

ArgyllCMS for example.


(Elle Stone) #38

Thank you @Morgan_Hardwood. I’m still not sure which one of the many images with targets actually has the colorchecker 24 and which “ev” to use.

Also my eyesight isn’t very good. Staring at that long list of thumbnails shot with a lens that I don’t actually have, to see which one might be the right one to download, to see if it might be useable with a reference file in ArgyllCMS, that might or might not have reference values that actually correspond to the particular target in the shot, to make a lut profile using a mere 24 color patches, seems like more trouble than it’s actually worth.

I do see an it8 chart in one set of thumbnails, but it’s rather small and off to one side, which isn’t how I’d shoot a target chart.


(Elle Stone) #39

Oh, that’s exactly the miserable mess I got out of dcraw (and other raw processors) when trying to shoot my favorite dark blue glass cup when light is shining through it. Which eventually led to outputting raw color, making a duplicate, assigning a matrix profile to one copy and a lab lut profile (made from the same target chart shot as the matrix profile) to the other, and blending in GIMP using masks and layers to only pull in the “lut” colors for areas where the backlit blue colors went wonky.

The walls, chairs, green signs, and such look really nice, very believable. As you say, the blue looks a little desaturated. What does this profile do to blue skies? Or would you use the modified profile only with shots with problem blue colors?

Edit: actually, thinking about it, in my photographs the backlit blue is just that, the result of backlighting various objects using non-blue light. In your photograph the blue light is the main source of light for the entire scene. Very different situations.


(Glenn Butcher) #40

I cropped the bottom of the image to remove people who I don’t know. But in that region, there are other manifestations of the change in the blue vector. One thing stymieing me is that the dominant illumination is definitely not sunlight, but all the profiles I have are based on sunlit targets. I am going to coordinate a visit to the theater, shoot a few shots of my colorchecker…

Hmm, looking at these images on my unmanaged work computer, they appear darker than they did last night at home. You’re managed, so you should be seeing what I saw, but the blue lights are accents, and the overall illumination is “normal”, so to speak.

I’d love ‘one profile to rule them all’, but I don’t know how close a LUT profile will get me in these situations. I think the best I’ll eventually get is a specific profile that handles the extreme blue case while not goofing up the rest of the image too much.