[argyllcms] Re: Difficulty shaping Black ink curve for CMYK profile
- From: Samuel Chia <inksandpaper@xxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Wed, 19 Dec 2018 13:49:44 +0800
Without specific examples, nothing like that is obvious to me.
Here is a TIFF file showing a gradients target with the original version on
the left and the one on the right was converted using Rel Col with BPC. In
particular, note the two middle rectangular granger rainbow gradients,
which show tonal reversals in the bands, getting lighter then darker.
https://drive.google.com/file/d/1NINzPyeC1vcWNvu_7_DWn0bV3nPFTMNV/view?usp=sharing
In what way is it affected by the ink limit setting ? - your example only
used 400%.
Sorry, no, my example uses 350%. I've only just re-discovered this issue I
noted earlier but forget to mention when making screenshots to provide
evidence to my claim. Although the ink limit was set to 350% during the
profile build (the commands used were: colprof -v -qh -kp 0 0.33 0.95 1 0.8
-l350 -tlp -s 20 -cmt -ca:27 -cg:15) from my inspection of the profile in
ColorThink, I see that the total ink goes from 350% at around input L*12,
up to around 362.5% at around L*8, before returning to 350% at dmax. Please
see the screenshots here:
https://drive.google.com/file/d/1ns4wNu6MT_uIMolwvPHlCWfRqQLI7zYI/view?usp=sharing
Hey, maybe it's this inking bug that is the cause of the curves issue!
Standard tiff is limited to 2GB or less, but if your files are less than
that,
I'm not aware of any problems with cctiff and large files, unless libtiff
itself simply
runs out of memory.
I think TIFF files can go up to 4GB. Photoshop is able to open and save
such files. My files frequently exceed 2GB and even 4GB, and I have to use
the PSB format.
Since I'm not privy to the workings of i1Profiler, or even have easy access
to
profiles created with it, I can only speculate that it goes about things in
a different way.
If there were any simple improvement to ArgyllCMS in this area that
I'm aware of, then I'd make it. Maybe someone with access to i1Profiler
could
do the necessary technical investigation of the contents of their B2A tag
to
give me the essential hint.
I am happy to provide the profiles, if you are interested. Here are the
i1Profiler profiles built from the same measurement set, the only
difference being black curve generation. As you can see, the black curves
in some instances are happy to be concave when approaching dmax, unlike
ArgyllCMS.
https://drive.google.com/file/d/1a74aQc9sO2TyGnR21J2ehxt6GwcOME88/view?usp=sharing
I've already provided the ti3 file in the previous thread but I'll copy it
here.
https://www.dropbox.com/s/txarthlw4ksvfgp/KHLIndigo10000-Nevia-limit400-CalGlossyDefaults-31Aug18-SpectralM0-averaged-kh-s20-ca27-cg7.ti3?dl=0
I'm not going to be the best person to do such a technical investigation,
although I am able to look at the visual results of gamut mapping and tell
what's wrong with it. So hopefully these additional data points are of some
help. I cannot say if it's an easy fix (maybe the ink limit issue is the
easy fix?), but it does seem to point to something perhaps more problematic
lurking in the code. Of course I am likely to be totally wrong in this
regard, so please forgive any transgressions.
I understand Photoshop can use device links.
I shall copy Doug's excellent (and more informed than I am) thoughts on the
matter. In addition, having to build the source gamut each time for
building the device-link profile is not efficient in a workflow of
thousands of images, especially in book production. The regular printer
profiles are actually not problematic at all in real world use. I have
never ever faced the need for the precision of device-links when converting
regular RGB photographs to RGB printer profiles, or even CMYK profiles.
Only now ArgyllCMS is giving me this problem with the tone reverals and
black ink curve shaping, otherwise I'm perfectly happy to use said
profiles, once the luminance-preserving gamut mapping is improved.
I'm running Win 10 x64, Photoshop CC.
While Photoshop can use device link profiles for converting from one device
to another this is typically used for making high quality Rel Col mappings
or remapping gamut in Perceptual. It's great for both of those but then you
are in device space. Now what? If the goal is to print it w/o color
management (it's already been done) Photoshop strongly discourages
bypassing color management via null transforms and the only other way to
work with device space RGB according to Photoshop is to first convert back
to PCS (LAB). So device link profiles can be used in conversion but the
results can't be printed directly. Just nutty.
However, on Windows the null transform has always worked flawlessly but
apparently not on Macs. So I don't see device link profiles as a viable way
to effect a null transform on Macs.
From:
https://forum.luminous-landscape.com/index.php?topic=122697.0
But short of ArgyllCMS miraculously becoming financially
self sustaining, I don't currently have the leisure to investigate this in
the depth needed to solve this limitation, particularly when it can be
avoided
by the superior device link workflow.
I've made the appeal to you several times privately and once more publicly
on the ArgyllCMS list that getting the luminance-preserving gamut mapping
to work correctly would be key to providing a solution that none of the
other industry players are capable of, nor as any printmaker ever in
history had the luxury of such beautiful control. Because many
photographers don't understand (have never seen) how lacking the current
choices of gamut mapping is, and how little visual shifting their images
would undergo with better gamut mapping, they are not yet demanding for
such a product. At least one luminary printmaker, Joseph Holmes, is very
interested in ArgyllCMS once this is figured out, and there are several
others who'll be very interested too. I've even offered to come down to
Australia to work with you in person towards this. Of course, there's no
guarantee that this will suddenly make Argyll a financial success, some
risks have to be taken in every endeavour, but at the current state, I
can't recommend it to anyone else - BUT I really want to. I can only do so
much to appeal to you, and I sincerely hope you will consider it
optimistically. I care a lot about this, and others can be educated to care
more too.
Sincerely,
Samuel
On Wed, 19 Dec 2018 at 13:07, Graeme Gill <graeme@xxxxxxxxxxxxx> wrote:
Samuel Chia wrote:
That's interesting. I actually see tonal reversals in gradients which
was why I didn't
assume it was only interpolation limitation.
Without specific examples, nothing like that is obvious to me.
It also seems to be greatly affected by the profile ink limit setting. I
suppose that's
tied to the interaction with the device response.
In what way is it affected by the ink limit setting ? - your example only
used 400%.
This profile was already built with "-qh" (33 CLUT points). I'll
certainly try "-bu". Is
that the same as using "-qu"? How many CLUT points would it create?
-bu is a way of setting a B2A table resolution that is independent of the
A2B table.
I think -bu is 45.
Unfortunately, as I have mentioned to you before, specifying the source
gamut and using
the device link workflow you taught me still does not allow me to fully
specify the
luminance vs saturation tradeoff for gamut mapping.
I'm not aware of anything that colprof can do that collink can't in that
area.
In addition, ArgyllCMS crashes with
the large TIFF files that I work with to make large prints.
Standard tiff is limited to 2GB or less, but if your files are less than
that,
I'm not aware of any problems with cctiff and large files, unless libtiff
itself simply
runs out of memory.
I am thus prevented from using
this otherwise higher precision method.
I understand Photoshop can use device links.
It's also strange that the i1Profiler profiles
never suffer from this Mach banding interpolation issue, but I've seen
it rear itself with
Argyll for both RGB and CMYK printer profiles. Makes one wonder why.
I've never seen this problem to the same degree in RGB profiles.
Since I'm not privy to the workings of i1Profiler, or even have easy
access to
profiles created with it, I can only speculate that it goes about things in
a different way. For instance, if you aren't concerned about the device
behavior being non-monotonic, don't mind loosing some accuracy, and don't
mind
loosing some control over clipping behavior, you could apply an ink rule
to the
CMYK values and simply treat the B2A PCS -> CMYK the same way as the A2B
table,
and model it with scattered point grid fitting. Or maybe they have some
hand
crafted per channel input and output curves. Or maybe they better fit
the expected PCS source gamut to the table. Or something else more clever,
I don't know.
If there were any simple improvement to ArgyllCMS in this area that
I'm aware of, then I'd make it. Maybe someone with access to i1Profiler
could
do the necessary technical investigation of the contents of their B2A tag
to
give me the essential hint. But short of ArgyllCMS miraculously becoming
financially
self sustaining, I don't currently have the leisure to investigate this in
the depth needed to solve this limitation, particularly when it can be
avoided
by the superior device link workflow.
Cheers,
Graeme Gill.
Other related posts:
