Luminous Landscape Forum
Raw & Post Processing, Printing => Colour Management => Topic started by: sgwrx on August 02, 2015, 03:24:26 pm
-
hello. i did my first paper custom profile using argyllCMS and colormunki photo. i followed this excellent write up: http://www.ludd.luth.se/~torger/photography/argyll-print.html. i printed 420 patches on 2 A4 sheets and my results are good. i have to get my head around this though.
paper: canson infinity baryta photographique
printer: epson 3880 with K3Vm inks
printing from photoshop with Relative Colormetric and BPC
i'm shocked at the difference between canson's profile and my profile when printing a test page from Andrew's website (2014 Printer Test File | A TIFF RGB 16-bit Adobe RGB (1998) test file (NEW).)
1. my profile appears more quite a bit (i'd say a lot) more vibrant in colors, especially in the Y C M gradients. The B looks about the same, the G looks a little more vibrant and a little darker, the R is just plain more red.
2. my profile has more yellow-green colors and olive greens whereas the canson has almost none.
3. my profile's "darker tones/shadows" appear darker overall but i can still see detail. for example in DD's image, on the lower left corner there is a small chart from blacks to lights right above the word "Date". the prints look similar and i can see colors in the patches, but as you look at the canson profile print it's like the "brightness" was turned up a tad.
4. my profile's dark blues and dark blue/magentas do not look as vibrant as the canson's. but all other dark saturated colors still have color and again appear more vibrant, but less than mid and lighter colors.
5. many of the colors on the canson profile print just look like one big color.
i don't know how to ask this, but is this most people experience when making their own paper profiles? to me, it's a good thing. i like it better. but i'm not looking for "it's your own personal opinnion".
i'm also surprised because my .icm file is 1/2 the size of canson's. partly due to using only Rel Col when building the profile within Argyll, but even with the other rendering intents by including the AdobeRGB profile at creation time, the profile only increased about 20%.
finally, as i've said i only used 420 color patches on two A4 sized prints. perhaps this is too few and i'm getting something that is not accurate?
*all the differences i see in print i also see on-screen when soft proofing in PS.
thanks for your responses. i apologize if this general idea has been discussed before. i've read many things for a long time and when you actually finally do it, it just seems different than what you read about other's experiences :)
-
i don't know how to ask this, but is this most people experience when making their own paper profiles? to me, it's a good thing. i like it better. but i'm not looking for "it's your own personal opinnion".
i'm also surprised because my .icm file is 1/2 the size of canson's.
You prefer the custom profile, that's the idea ;D
You could surmise that the canned profile isn't so good (not a surprise)
Size doesn't matter, I keep telling my wife that! :D There could be a lot of metadata inside the profile that add's to it's size.
-
The number of patches depends on how well-behaved is the printer. For a CMYK press I may use large numbers of patches. Current desktop printers are very well-behaved (I would say Epsons are such for 5 years already at least), so I do not think you a loosing much with 420 patches, especially if you used pre-conditioning. You may find that 2 pre-conditioning steps over 210 patch targets result in more improvement than changing to a 1500+ patches final target. You do not need to re-print and re-measure pre-conditioning targets when you change the ink or open another batch of paper; you just re-use the profile created on the pre-conditioning stage.
Another factor that determines the amount of patches necessary is the paper. For a good paper the above stands, but if the paper is not designed to take the particular ink well, more patches may be needed. To check, I simply print wedges after profiling and check linearity. Free ColorPort is actually a very useful tool to create such patch sets and render those.
To check error numbers, I first print ColorChecker SG and measure it. If the max deltaE2000 is less than 3, in my experience not much can be done to improve the printer profile.
-
thanks both. i know there's a ton of influences, but i've never seen anyone remark that they were shocked at the differences.
i guess this is the point where i pass judgement on my deliberation and purchase of the colormunki photo, it was worth it. plus i needed to ditch my aged i1d2.
-
Free ColorPort is actually a very useful tool to create such patch sets and render those.
Unfortunately, X-Rite has discontinued ColorPort
ColorPort Utility Software (Discontinued)
We recommend i1Publish as a replacement for ColorPort Utility Software (Discontinued).
https://www.xrite.com/colorport-utility-software/support/d1168 (https://www.xrite.com/colorport-utility-software/support/d1168)
I've been experimenting with ArgyllCMS targen and printarg but haven't figured out how to make them generate ColorChecker (or similar) clone charts. I tried printing the Dry Creek Photo electronic versions of the GretagMacbeth ColorChecker charts (https://www.drycreekphoto.com/Learn/monitor_calibration.htm#charts), but they aren't matching very well (my best matches are by setting Photoshop to "let printer manage colors" and set the printer driver to full auto.)
I was testing 392 patch Argyll profiles. And profiles made with X-Rite ColorMunki Design. All made with a new (several months) ColorMunki. I printed using both relative colorimetric and perceptual. Prints from each rendering intent had one or more wonky colors (DeltaE2000 > 12.) But not the same colors. Most other colors had Delta Es in the 5-7 range. I was testing by using ArgyllPro ColorMeter and comparing patch by patch against a real ColorChecker Passport. (I couldn't figure out how to do this test with Argyll chartread. I have figured out how to use chartread to use the supplied colorchart.ti2 file so I can measure CC(ish) charts, but I don't know enough color science math to write the DeltaE algorithms in a Perl script that would compare .ti3 files that chartread generates. I can read and parse CGATSish files with no problem, but I am lost with math-to-algorithms. (i.e., http://www.brucelindbloom.com/index.html?Math.html (http://www.brucelindbloom.com/index.html?Math.html)))
Wayne
-
Unfortunately, X-Rite has discontinued ColorPort
ColorPort Utility Software (Discontinued)
So is MeasureTool. Doesn't mean it doesn't run under the proper OS and produce the measured data one desires from it. Same with ColorPort.
i1Publish isn't free.
-
So is MeasureTool. Doesn't mean it doesn't run under the proper OS and produce the measured data one desires from it. Same with ColorPort.
i1Publish isn't free.
I still use ColorLab! :)
The lack of decent colour target editors supported by modern OSes is unfortunate, though to make one is pretty simple - but still takes time.
-
So is MeasureTool. Doesn't mean it doesn't run under the proper OS and produce the measured data one desires from it. Same with ColorPort.
Yes, but this knowledge isn't useful if the programs aren't available. Which is why I was asking about how to do the same thing with Argyll programs (or Perl algorithms.) Argyll (and color science math) are really powerful, but are sometimes difficult to comprehend. Even after reading the excellent instructions at http://www.ludd.luth.se/~torger/photography/argyll-print.html (http://www.ludd.luth.se/~torger/photography/argyll-print.html).
As you have noted many times, XRite makes excellent hardware but their software leaves a lot to be desired. Perhaps we should walk away from XRite software and embrace Argyll. But some of us need help understanding how to make Argyll dance.
Wayne
-
I still use ColorLab! :)
The lack of decent colour target editors supported by modern OSes is unfortunate, though to make one is pretty simple - but still takes time.
I've got on an very old Mac too, it rocks. And yes, it's a shame there are so few tools to edit targets. Sing of the times I guess.
-
Yes, but this knowledge isn't useful if the programs aren't available.
MeasureTool and ColorPort are both available.
-
I've got on an very old Mac too, it rocks. And yes, it's a shame there are so few tools to edit targets. Sing of the times I guess.
10.6.8 - runs without issues. For more recent OS X versions - Windows ColorLab version and Crossover, smooth.
The 'unfortunate' thing is that I learned to edit CGATS files with available tools manually (ColorLab, ColorPort, Excel, text editor), and have no immediate motivation to build an editor, while having lots of other things to do...
-
Yes, but this knowledge isn't useful if the programs aren't available. Which is why I was asking about how to do the same thing with Argyll programs (or Perl algorithms.) Argyll (and color science math) are really powerful, but are sometimes difficult to comprehend. Even after reading the excellent instructions at http://www.ludd.luth.se/~torger/photography/argyll-print.html (http://www.ludd.luth.se/~torger/photography/argyll-print.html).
As you have noted many times, XRite makes excellent hardware but their software leaves a lot to be desired. Perhaps we should walk away from XRite software and embrace Argyll. But some of us need help understanding how to make Argyll dance.
Wayne
Takes time and practice; it is not that an excellent profile can be obtained by some magic combination of commandline parameters. Argyll mailing list is extremely helpful if you need your questions to be not just answered but discussed in depth. But of course you can try asking here as well.
-
MeasureTool and ColorPort are both available.
ColorLab https://www.xrite.com/product_overview.aspx?ID=1071&Action=support&SoftwareID=486
-
Takes time and practice; it is not that an excellent profile can be obtained by some magic combination of commandline parameters.
I know that it takes time and practice. I've been reading color science textbooks, along with crawling the web for color related material. Color science is complicated such that it looks to me like it requires a serious formal education before you can write tools from scratch.
But a cookbook approach is a way to bring tools to ambitious amateurs. I don't have any formal education in computer science. But I can get a lot accomplished with some basic programming knowledge and books like the Perl Cookbook (http://www.amazon.com/Perl-Cookbook-Second-Edition-Christiansen/dp/0596003137)
The ArgyllCMS site has the beginnings of an Argyll Cookbook in the Typical usage Scenarios and Examples (http://www.argyllcms.com/doc/Scenarios.html) section. But...
Argyll mailing list is extremely helpful if you need your questions to be not just answered but discussed in depth.
I've been subscribed to this list for the past several months. But it isn't geared towards answering the kinds of questions that are better answered in a cookbook (a typical "recipe" is longer than would comfortably fit in a single email.)
ColorLab https://www.xrite.com/product_overview.aspx?ID=1071&Action=support&SoftwareID=486 (https://www.xrite.com/product_overview.aspx?ID=1071&Action=support&SoftwareID=486)
Thanks. I didn't know about this. I installed it and am exploring it. I see that it can read Argyll .ti3 files (CGATS variant) but can't read ColorMunki Design .cxf palette files. Interesting.
It looks like loading .ti3 files that were made by chartreading (http://www.argyllcms.com/doc/chartread.html) different CC charts (real XRite CC vs. CC charts printed by my printer with various profiles, etc.) and using ColorLab Tools/Compare lets me compare the Delta Es for different chartread measurements. Cool. Now I wonder if there is a way to use any of the Argyll programs to do the same thing?
The 'unfortunate' thing is that I learned to edit CGATS files with available tools manually (ColorLab, ColorPort, Excel, text editor)
What do you mean by "edit"? CGATS files are simple ASCII files that can be easily decomposed into key/value pairs and/or list/array/hash fields with Perl regular expressions (or the equivalent in whatever language you prefer.) It is the manipulation of the data that is difficult (for me), because it requires knowing color science math (http://www.amazon.com/Color-Science-Concepts-Quantitative-Formulae/dp/0471399183).
I could, for example, write a Perl script to read two .ti3 files and parse them into variables. But I don't know how to do the algorithms that duplicate what ColorLab does with Tools/Compare to generate Delta Es. But I could if there was a cookbook recipe (math to algorithm) for this particular operation.
Wayne
-
I suppose you could try using 'swig' to wrap littlecms (which has various dE functions) so that you can call it from PERL.
swig.org
http://www.swig.org/Doc3.0/Perl5.html#Perl5
I haven't looked at .ti3 but if it outputs Lab colors, the dE76 equation is super-simple (and 94 isn't hugely difficult either)
and if you want to get really cheesy about it, write the code so it goes out to the web and uses one of the online color libraries.
-
Some ways to compute deltaE:
http://www.argyllcms.com/doc/colverify.html
Another useful, but not free, tool - http://www.babelcolor.com/main_level/PatchTool_features.htm (free pdf doc goes into detail)
Yet another tool we use, http://www.chromix.com/colorthink
And one more, http://www.rmimaging.com/spectrashop.html
Equations at https://en.wikipedia.org/wiki/Color_difference
If you are OK with Python, there is a free library https://pypi.python.org/pypi/colormath/ with http://python-colormath.readthedocs.org/en/latest/delta_e.html
As to editing targets, it is easier when one has a WYSIWYG editor to see layout of the target and to add/delete particular colours and colour gradients.
-
I'm really surprised there is no PERL CPAN module.
-
Preconditioning? Multiple preconditioning steps? How would one go about this? I use i1Profiler for RGB profiling. I just want a good idea on how I would go about this. As far as the editing of CGARS files, I use an regex style editor that has a GUI interface.
-
Preconditioning? Multiple preconditioning steps? How would one go about this? I use i1Profiler for RGB profiling
I don't know how to do it in i1Profiler. Preconditioning was discussed in Anders Toger's Inkjet printer profiling with Argyll and Colormunki (http://www.ludd.luth.se/~torger/photography/argyll-print.html) tutorial that we have been referring to in this thread.
Skim it, but pay special attention to the Target generation details section, which explains why preconditioning may allow for a better choice of target patch colors.
I am just learning Argyll. I have never used i1Profiler to profile and calibrate printers. I would really like to read a comparison of Argyll and i1Profiler, done by somebody that knows color science well.
Wayne
-
i1P supports what could be called 'post-conditioning' or just optimization. It generates an X patch target and after printing and scanning, you can 'optimize' the profile by having i1P generate another target that zeros in on parts of the space where it may think things are in need of additional tuning and data points.
Seems like Argyll does it exactly opposite by generating a pre-conditioning target and then you feed those results into the main target generation.
-
I don't know how to do it in i1Profiler. Preconditioning was discussed in Anders Toger's Inkjet printer profiling with Argyll and Colormunki (http://www.ludd.luth.se/~torger/photography/argyll-print.html) tutorial that we have been referring to in this thread.
Skim it, but pay special attention to the Target generation details section, which explains why preconditioning may allow for a better choice of target patch colors.
I am just learning Argyll. I have never used i1Profiler to profile and calibrate printers. I would really like to read a comparison of Argyll and i1Profiler, done by somebody that knows color science well.
I was wondering what you guys were talking about.
I'd call the first step linearization, or at least a step used to build targets for a profile was historically called that. Maybe preconditioning is different?
I'd call the last step optimization, or at least a step to produce a better profile after initial profile creation.
-
I was wondering what you guys were talking about.
I'd call the first step linearization, or at least a step used to build targets for a profile was historically called that. Maybe preconditioning is different?
I'd call the last step optimization, or at least a step to produce a better profile after initial profile creation.
The idea of the "pre-conditioning" is to get a rough idea of the gamut and not to waste patches on sampling out-of-gamut; but rather, knowing the response, to place targets in such a way that the existing gamut is sampled better. Ideally "knowing response" also is "accounting for non-linearities", of course.
From help/manual on Argyll:
"The -c flag and parameter is used to specify an ICC ... pre-conditioning profile, for estimating perceptual distances and colorspace curvature, used in optimizing the full spread test point placement, or in creating perceptually spaced distributions. Normally a previous profile for this or a similar device will be used, or a simpler, preliminary profile will be created and used."
"To assist the determination of test patch values, it can help to have a rough idea of how the device behaves, so that the device test point locations can be pre-conditioned. This could be in the form of an ICC profile of a similar device, or a lower quality, or previous profile for that particular device. If one were going to make a very high quality Lut based profile, then it might be worthwhile to make up a smaller, preliminary shaper/matrix profile using a few hundred test points, before embarking on testing the device with several thousand."
-
i did do the pre-condition, even though i don't fully understand it from the documentation. but my question is: what if the profile you started with is very poor quality? i guess that's when they say you can make a pre-condition target within Argyll comes in?
secondly, i'm not sure what's going on with the 2-step precondtion, but i'm guessing you run create a profile with precond set, then create another one using that second profile?
just a quick side question - you know the physical color checkers? i have a lalrge one that's about 6-7 years old, but then i just got a new one with the colormunki photo. my question is this: are the colors on these devices computer generated in LAB space with a certain lighting in mind like D50? then they have to do some uber-color control in order to print it out or create a paint like substance to actually create the card?
so it all starts in the computer then and then gets translated into the physical world?
-
> i did do the pre-condition, even though i don't fully understand it from the documentation. but my question is: what if the profile you started with is very poor quality?
The first, pre-conditioning, profile characterizes the borders of the gamut, and the device linearity. It allows for targen to follow something better than an abstract device model when generating patches for a final profile. Abstract model places the patches "all over the place", shooting in the dark. Pre-conditioning allows to put patches "on target", that is within the gamut of the device; and distribute those patches appropriately.
> i'm not sure what's going on with the 2-step precondtion
Just an improvement over the initial device description, plus if the first and second preconditioning profiles differ significantly, it is an early diagnostics of device instability or workflow problems (like not letting the target to dry fully, or wrong parameters in commandline).
> i'm guessing you run create a profile with precond set, then create another one using that second profile?
Yes.
> just a quick side question - you know the physical color checkers?
You mean, do I have those targets we shoot? Yes, I do.
> are the colors on these devices computer generated in LAB space with a certain lighting in mind like D50?
That is a long and very interesting story, you may want to start the reading with the bio of the man named Munsell, and his work of life. The colours are carefully selected from real-life colours and actually named after those colours, except some that are close to printing pigments (sort of real-life, too). The colours were coded in Munsell notation, and later converted to Lab (which is D50) for convenience.
> so it all starts in the computer then and then gets translated into the physical world?
From the above, you can see it is vice versa ;)
-
thanks!
oh, so when a patch is named "skin tone" or "foliage" it's matching something real! interesting, i'll do some reading tonight. thanks!
-
when a patch is named "skin tone" or "foliage" it's matching something real!
There are 2 patches on ColorChecker Classic (24 patches) for the skin tone. One is "darker than light skin", the other is "lighter than dark skin".
-
> i did do the pre-condition, even though i don't fully understand it from the documentation. but my question is: what if the profile you started with is very poor quality?
The first, pre-conditioning, profile characterizes the borders of the gamut, and the device linearity. It allows for targen to follow something better than an abstract device model when generating patches for a final profile. Abstract model places the patches "all over the place", shooting in the dark. Pre-conditioning allows to put patches "on target", that is within the gamut of the device; and distribute those patches appropriately.
Am I correct that a generic icc-profile from that paper/printer, for example from the manufacturer, could be used for the preconditioning?
-
Am I correct that a generic icc-profile from that paper/printer, for example from the manufacturer, could be used for the preconditioning?
yes. according to manual even something from a similar paper. i used the generic photographique profile. i bet i could have used ilford gfs profile.
Sent from my XT1080 using Tapatalk
-
Am I correct that a generic icc-profile from that paper/printer, for example from the manufacturer, could be used for the preconditioning?
I would not do that, it is a dubious shortcut. Some canned profiles are so far off that they do more harm than good.
-
I would not do that, it is a dubious shortcut. Some canned profiles are so far off that they do more harm than good.
Are the canned profiles so bad or are the printers still so different?
-
Are the canned profiles so bad or are the printers still so different?
Canned profiles are not very good. An error will cost 2x the work and time delays for drying, why take the risk.
-
> i did do the pre-condition, even though i don't fully understand it from the documentation. but my question is: what if the profile you started with is very poor quality?
The first, pre-conditioning, profile characterizes the borders of the gamut, and the device linearity. It allows for targen to follow something better than an abstract device model when generating patches for a final profile. Abstract model places the patches "all over the place", shooting in the dark. Pre-conditioning allows to put patches "on target", that is within the gamut of the device; and distribute those patches appropriately.
so what am i mis-understanding about pre-conditioning? if you do not use (recommended to not use) a canned profile from the paper mfg. (aren't epson's supposed to be pretty good?), then do you:
1. create a target w/o pre-conditioning, print it, measure it, create a profile (abstract device model as you say)
2. use this profile as pre-conditioning
?
i guess that makes more sense. but since it's a characterization of the borders of the gamut, wouldn't pretty much any profile work?
-
> if you do not use (recommended to not use) a canned profile from the paper mfg. (aren't epson's supposed to be pretty good?), then do you:
> 1. create a target w/o pre-conditioning, print it, measure it, create a profile (abstract device model as you say)
> 2. use this profile as pre-conditioning
Yes. Epson profiles are supposed to be good, and they are good; but in practice, I specifically checked that, creating a pre-conditioning profile from the scratch works better. However please keep in mind, my spectrophotometers are automatic, so I do not care how many patches I need to measure. Another thing with me is that I seldom go for shortcuts. I do not want anybody to get side-tracked into splitting hairs; just describing how I do it.
-
ah ok, makes more sense to me. thanks! so yeah, if the profile is bad it'd not be as good. so the best approach is to "shoot in the dark" as far as which patches are made automatically, then use that to hone-in.
i'm probably going to try 800 or so patches this week or next just to see how it goes for myself. i'll do it this way for preconditioning. that'd be 4 a4 sheets of paper. though i thought i read somewhere i could print the same size patches on 13x19 paper and get more of them on that paper. i'll read up on that.