or Connect
AVS › AVS Forum › Display Devices › Display Calibration › MadVR - ArgyllCMS
New Posts  All Forums:Forum Nav:

MadVR - ArgyllCMS - Page 27

post #781 of 1700
 
Quote:
 @Graeme
It would be very nice if we could only take the color space form the reference Rec709.icm file (without the embedded Gamma), and take the Gamma from the Measured.icm.
Originally Posted by gwgill View Post


It doesn't work that way. The whole idea is to obliterate the native behavior of the display, and to make it behave in an ideal way defined by the source profile + any modification such as BT.1886 + gamut mapping.

 

Yes, but the Rec709.icm file has a gamma of 709 which as we all know is for encoding only.

What is the point of emulating the gamma of Rec709 for an Electro-Optical device?

Therefor the result (if not using bt.1886 options) is way too bright.

 

Is there no way to merge my .cal file with the Rec709 color space and create .icm file with my gamma curve AND 709 color space?

 

I just don't see why do we need the gamma curve from other .icm files if we have our own calibrated one.

By default collink enforces the gamma curve from the Rec709.icm profile, which is not what any other color managed software does.

 

 

I want to create a 3DLUT that uses MY calibration (gamma) file, and shows perfect Rec709 colors.

That's it.

 

I want a 3DLUT that makes MadvR "Color Aware" but not play with my gamma in any way.

Until now there is no such option with collink or madvr at all.

 

Just like what MPC-HC does with Color Management natively.

 

Hope this makes it clearer.

 

Thanks.


Edited by James Freeman - 10/7/13 at 7:40am
post #782 of 1700
Quote:
Quote:
 Originally Posted by James Freeman View Post

collink.exe  -v -3m -qh -et -Et -G -a "Measured.cal"  -iaw  Rec709_gamma22.icm  "Measured.icm"  "3DLUT.3dlut"
collink.exe  -v -3m -qh -et -Et -G -a "Measured.cal"  -iaw  "Measured.icm" "Measured.icm"  "3DLUT.3dlut"
Originally Posted by gwgill View Post


You're likely to get crushed blacks with both of those (unless you have a very good black point for your display), since nothing is mapping the black point. -IB/b maps the black point, as does -ila etc.

 

Exactly !!!

 

I get elevated blacks with Rec709.icm

I get crushed blacks with Rec709_Gamma22.icm.

I get Blue whites with Measured.icm (No crushed blacks. Perfect gamma as I want it).

I get low contrast with BT.1886 (-IB/-Ib).

 

If only I could use MY calibration file with Rec709 color space... ;) 

Is there any way to embed and overwrite my .cal file into the Rec709.icm?

 

EDIT:

-ila did the trick.

Thanks.


Edited by James Freeman - 10/7/13 at 12:17pm
post #783 of 1700
I (almost) found what james Freeman wants:

collink -v -3m -qh -et -Et -G -ila -a"measure".cal Rec709_gamma22.icm "measured".icm "measuredLUT".icm

This gives no BT1886 processing, but in my case colors are not perfect (I get DeltaE up to 2 or 3 where there are errors below 1 using iaw or ir)

Blacks are perfectly...crushed, up to 4 IRE, still the image may be better, it's a matter of taste. Because there are less shadow details but in my case I get rid of a greenish color cast, which must be the outcome of the attempt BT1886 tries to do in order to improve shadow details.

think that poor displays, like most LCD TV, may give a better image even with black crushed rather then trying to make them distinguish low level details, but not because BT1886 is not a good or smart algorythm, but because LCD technology is really crap for video materials...
post #784 of 1700
Again about the elevated black point:
Wouldn't it be possible to edit the 3dlut in such a way that it maps black (16,16,16) to 16,16,16 again? Obviously it's mapped to something else in my 3dluts, otherwise I wouldn't see dithering on the TV.
My common sense tells me, that there has to be some line in the lut which says something to the effect of
Code:
16,16,16 -> 18,18,18

How can I look into the 3dlut? All my programs just show gibberish.
post #785 of 1700
Quote:
Originally Posted by Kukulcan View Post

I (almost) found what james Freeman wants:

collink -v -3m -qh -et -Et -G -ila -a"measure".cal Rec709_gamma22.icm "measured".icm "measuredLUT".icm

YES !!!!

Almost there.

 

Best setting so far.

 

EDIT:

I upped the resolution to -r150 which is way more than what -qu uses.

It looks identical gamma-wise, as without the 3dlut at all (its a good thing).

 

 

Quote:
 la - Luminance matched Appearance

Right under my nose, Go figure.

 

 

Thank you.


Edited by James Freeman - 10/7/13 at 8:32am
post #786 of 1700
Quote:
Originally Posted by James Freeman View Post

I upped the resolution to -r150 which is way more than what -qu uses.
I think madVR LUTs are always 65x65x65 regardless of the setting (could be wrong though).
post #787 of 1700
Quote:
Originally Posted by gwgill View Post

Quote:
Originally Posted by VerGreeneyes View Post

Edit: Holy crap, the 3DLUT created from an -as profile looks so much better. Black is actually black now! Thanks for the tip!

I suspect the result (especially near black) is rather dependent on the quality of the display profile, and that the current Argyll profiling defaults aren't putting sufficient emphasis on characterizing or modelling the display near black behaviour, making it a little hit or miss, particularly for displays with very low black points. It's something I'm intending to take a look into at some stage.
I'm not sure the black point of my display is particularly low, but Argyll seems to set it to something really high after profiling (somewhere on the order of 40/256 as far as the digital output goes). It also does this after calibration unless I use something like -g2.2 -a64, and I think what I'm seeing in the 3DLUTs is the profile trying to put back the high black point that I tried to avoid during calibration. So far the only combination of settings that has gotten me a 3DLUT with halfway decent black is the conceptually silly idempotent transform that James Freeman suggested - if I then add -IB:2.2 or -Ib:2.2 to that, the black point gets raised again.

Mind you, the problem isn't so much how much it raises black, but more the fact that anything near black (which, with any level of compression tends to be a blocky mess) becomes incredibly visible and grainy. Crushed blacks are one thing, but this is way too far in the opposite direction!

Edit: Sorry about double posting, that was my bad.

Edit2: Aha! I get a good result with the following:
Code:
collink.exe -v3 -3m -qu -r65 -G -Ib:2.2 -iaw -et -Et Rec709_Gamma22.icm Profile.icm 3DLUT.icm

This has the same black level as:
Code:
collink.exe -v3 -3m -qu -r65 -G -iaw -et -Et Profile.icm Profile.icm 3DLUT.icm
but looks somewhat more vibrant. I think we have a winner! (incidentally, I see no difference between -ir and -iaw with these command lines)

Purely out of curiosity, is there any potential advantage or disadvantage to setting a higher resolution that is a multiple of 65? Will it average the obtained values together in some way or just pick the nearest neighbor?
Edited by VerGreeneyes - 10/7/13 at 10:37am
post #788 of 1700

 

I have Calibrated and Profiled my U2410 in the Wide Gamut setting (Standard).

As you can see the native gamut of the monitor encapsulates the whole spectrum of Rec709/sRGB.

 

With the command Kukuclan suggested I created a 3DLUT to get full spectrum Rec709 Colors and Gamma curve to my taste.

Exactly what I wanted.

 

Just in case anyone missed:

collink -v -3m -qu -et -Et -G -ila -a "measured.cal" Rec709_gamma22.icm "measured.icm" "measuredLUT".icm

 

 

IMO:

Yes, the blacks ARE compressed ("Crushed").

I've always used "Black Output Offset" to 100% (-f1) for calibration which is also the Default setting in Argyll (even if you dont type -f at all). 

Other software do this by default (Power Law + BLC).

I can still see all the black tones from 1 and on, they do not clip or cut off, just compressed.

 

Any setting that tries to make more visible blacks like BT.1886, sRGB, or "Input Offset" (-f0), just destroys the already poor 

LCD "Visible" Contrast Ratio, and makes it less "Deep".

 

I  think I am very used to see Compressed Blacks on LCD, so moving to BT.1886 just looks too... black elevated and less deep.


Edited by James Freeman - 10/7/13 at 12:19pm
post #789 of 1700
@James, I think our monitors are a bit different or we just want different things, but I'd be curious to hear what you get when you use the commandline I posted above.

@Graeme: Now that I've found a set of parameters that work for me, I was wondering: does it make conceptual sense to use a perceptual mapping with collink? When I replace -ir or -iaw with -ip above I get a somewhat different result - a bit less contrast in places, either less vibrant or more subtle - but I can't visually decide which is better, so I was wondering if there's an objective reason to prefer one over the other.
Edited by VerGreeneyes - 10/7/13 at 10:56am
post #790 of 1700
Quote:
Originally Posted by gwgill View Post

How much is "too bright" ?

So much brighter

g 2.2 without argyll



with argyll
Code:
dispcal -v -d madvr -yn -XWLEDFamily_07Feb11.ccss -qm -w 0.3127,0.3290 -g2.2 -o TVmtx.icm TV

targen -v -d3 -s30 -g100 -f1000 -cTVmtx.icm TV

dispread -v -d madvr -yn -XWLEDFamily_07Feb11.ccss -K TV.cal TV

colprof -v -qh -bl TV > 4.colprof.log

collink -v -3m -et -Et -Ib:2.4 -G -ir -a TV.cal Rec709.icm TV.icm 7_HD.icm



Do you have any idea what's going wrong for me?
post #791 of 1700

@VerGreeneyes

 

Quote:
collink.exe -v3 -3m -qu -r65 -G -Ib:2.2 -iaw -et -Et Rec709_Gamma22.icm Profile.icm 3DLUT.icm

The blacks looks very crushed (no BLC) so it does not use your calibration file (it uses the Rec709_Gamma22.icm built in gamma curve).

Same as without -Ib.

 

VarGreeneyes, try the command Kukulcan seuggested.

 

 

@MSL_DK

 

Nothing is wrong, on the contrary, everything is right.

BT.1886 is way "black elevated" than your typical "Power Law + BLC" curve.

 

As already stated, If you don't like how BT.1886 looks, just stick with Power Law + BLC (what you got now).

Its what we always saw for years, we got used to it.

 

Tryy the command Kukulcan seuggested.

post #792 of 1700
Quote:
Originally Posted by James Freeman View Post



@MSL_DK
As already stated, If you don't like how BT.1886 looks, just stick with Power Law + BLC

And how do I do it with argyllcms?
post #793 of 1700
Quote:
Originally Posted by MSL_DK View Post


And how do I do it with argyllcms?

Try this as the last command:

 

collink -v -3m -et -Et -G -ila -r65 -a "TV.cal" "Rec709_Gamma22.icm" TV.icm "HD.3dlut"

 

Use this instead of Rec709:

Rec709_Gamma22.txt 1k .txt file

Change .txt to .icm

post #794 of 1700
Quote:
Originally Posted by James Freeman View Post

The blacks looks very crushed (no BLC) so it does not use your calibration file (it uses the Rec709_Gamma22.icm built in gamma curve).
Same as without -Ib.
Strange, it looks very different for me depending on whether I include the -Ib:2.2 part. Without it, the blacks look very crushed, but with it they look fine.
Quote:
Originally Posted by James Freeman View Post

VerGreeneyes, try the command Kukulcan seuggested.
I did, it looked awful for me - very high black level with a purple sheen.

@Graeme: Why are the results between Rec709.icm and Rec709_Gamma22.icm so different, do you know? Inspecting the profiles, the only difference appears to be that Rec709.icm uses 4096 points representing the transfer function, whereas Rec709_Gamma22.icm uses the pure power function - but why would the former raise the black level and the latter leave it alone? (I can understand that the ramp is different near black, but why would black itself be raised?)
post #795 of 1700
Thread Starter 
Quote:
Originally Posted by VerGreeneyes View Post

Code:
collink.exe -v3 -3m -qu -r65 -G -iaw -et -Et Profile.icm Profile.icm 3DLUT.icm

Just to reiterate what Graeme has stated, when you link your measured profile with itself, no corrections will be made with the 3DLUT. wink.gif
post #796 of 1700
Quote:
Originally Posted by N3W813 View Post

Just to reiterate what Graeme has stated, when you link your measured profile with itself, no corrections will be made with the 3DLUT. wink.gif
I was just using that as an example of a 3DLUT that didn't raise my black level. I've moved on from that wink.gif
post #797 of 1700
I also have a Dell U2410, and have been using the 3dLut created by the first page of this post. Great color accuracy, but made the black levels unacceptable for the U2410, which is already way higher than most modern LCD displays. I tried this latest command line

collink -v -3m -qu -et -Et -G -ila -a "measured.cal" Rec709_gamma22.icm "measured.icm" "measuredLUT".icm

It gave me better blacks, before the black would be way beyond the reference black of 16, more like 13 or so, now it's around 15, so much better for me. What I still have to find out is if it affected the accuracy of the LUT, I will be comparing the two of them.

One question, it seems some users are using the DellU2410 in Standard mode, then using the 3dLUT to get the colors to within the 709 color space. Isn't it easier to use Dell U2410 own sRGB mode, which does it via hardware, which must result in a better quality of image, without any image artifact, than using software to do the same? Just curious.
post #798 of 1700
I wanted to experiment a bit with different transfer functions, so I fired up Mathematica and made some variants of Rec709.icm with different transfer curve tables. The results are in this Mediafire folder if you'd like to try them. In alphabetic order, they are:

Rec709-1.95.icm - Pure power function with a gamma of 1.95, the same effective gamma as BT.709.
Rec709-2.23.icm - Pure power function with a gamma of 2.23, the same effective gamma as sRGB (basically the same as Rec709_Gamma22.icm).
Rec709-BT.709.icm - The same as normal Rec709.icm, but with the discontinuity removed (to please my OCD).
Rec709-sRGB.icm - Uses the transfer function from sRGB (again with a small tweak to remove a discontinuity).

On my monitor, Rec709-1.95.icm comes out looking the best, followed closely by Rec709-2.23.icm (which is a bit too dark overall) and Rec709-sRGB.icm (which still looks a bit harsh near black). The linear segment of the sRGB transfer function is a lot more shallow than that of BT.709, so it doesn't end up looking too bad. Stock BT.709 is still way overkill as far as my monitor is concerned.
post #799 of 1700
@Alec246

Sometimes a typical sRGB monitor do not cover the whole sRGB spectrum (something like 95%).

The U2410 does covers some colors completely but a little short for others.
For example:
sRGB emulation mode covers 99% of sRGB spectrum.
Whether Wide mode covers 100% or more.
That's why I (Sometimes) use the Wide (Standard) mode to be sure that I cover the whole 100% sRGB spectrum.



@VerGreeneyes

Have you managed to implement a custom gamma curve in an .ICM file?
If so, Share..
post #800 of 1700
Quote:
Originally Posted by James Freeman View Post

Have you managed to implement a custom gamma curve in an .ICM file?
If so, Share..
Technically yes, it should be doable, though you'd have to do it by hand. I used the IccXMLTools to convert Rec709.icm to an xml file so I could edit it - you'd have to take the curves from your .cal file, convert them to the range 0-65535 and put them in for rTRC, gTRC and bTRC. I'll see if I can make an example (though I'm not sure whether or not this will work right).

Edit: Okay, I made an example and put it in the MediaFire folder. There you'll find Gamma.ods / Gamma.xls showing how to convert from the .cal format to the ICC format ([0-1] -> [0-65535] and rounded), and Rec709-Custom.xml which shows where to insert the curves. You'll then need to use IccFromXml.exe from the IccXMLTools to convert the xml back to an icc profile.

Whether this will work the way you want it to I don't know. The .cal curves are just corrections, so you might have to do some additional work to translate the data to the actual gamma measured for your monitor.

Now, it's not entirely clear to me what you think should be in the profile. Assuming your .cal file is loaded into your GPU's videoLUT, the intent is that the gamma (after applying the videoLUT) should match the theoretical curve that dispcal was shooting for. So if, for instance, you chose -gs for dispcal, isn't the transfer function you want simply the sRGB transfer function?
Edited by VerGreeneyes - 10/8/13 at 5:31am
post #801 of 1700
Graeme, we need your help...

is it possible to create a LUT without BT1886 processing and a pure power law gamma curve? If we don't put in IB/b in collink, we get weird LUTs, clearly faulty. The collink sequence I posted (with Rec709gamma22 and ila) it's faulty anyway, yes, it gives a pure gamma power law curve but colors are a bit wrong and I see artifacts near black.

I think that some people may prefer crushed blacks rather than the weird look that cheap lcd displays give when forced to distinguish near black levels and also up to 10-15 IRE.

Just for info, here below my measures for pure black from a video source signal (black screen in the movie Cars at the beginning), luminace Y is very consistent

native display: x:0.231 y:0.205 Y:0.088
profiled x:0.235 y:0.207 Y:0.093
LUT iaw and Ib x:0.233 y:0.208 Y0.093
LUT iaw but no IB/b x:0.313 y:0.326 Y:0.251 (numbers say almost D65, but it seems a terrible green!)
LUT ila Rec709gamm22 no IB/b x0.233 y:0.208 Y:0.093

the last one remains crushed at 0.093 up to 4 IRE, and then raises at a slower rate (of course since no BT1886), still it may be preferable...
post #802 of 1700
Quote:
Originally Posted by Kukulcan View Post

Graeme, we need your help...

is it possible to create a LUT without BT1886 processing and a pure power law gamma curve? If we don't put in IB/b in collink, we get weird LUTs, clearly faulty. The collink sequence I posted (with Rec709gamma22 and ila) it's faulty anyway, yes, it gives a pure gamma power law curve but colors are a bit wrong and I see artifacts near black.

I think that some people may prefer crushed blacks rather than the weird look that cheap lcd displays give when forced to distinguish near black levels and also up to 10-15 IRE.

I also missing this option. It would be great!
post #803 of 1700
Again, just for being clearer, please look at these numbers in my best LUT (for my LCD, not my Planar DLP...)

Black: x0.231 y0.208
1 IRE x0.244 y0.229
2 IRE x0.264 y0.257
3 IRE x0.281 y0.284
4 IRE x0.294 y0.300


They're so damn off from D65 that I say better not to see them at all! The darker they are, the better it is...
post #804 of 1700
Graeme has a nice discussion of the alternatives to BT.1886 black point compensation in section 8 and below here, using -ila appearance mapping with or without BT.1886 mapping. You may want to try some of those parameters. Also make sure you have added at least 8 (I use 16) black measurements to your targets to pin down the value in the profile. Even doing that I will run into cases where the LUTed black point will be mapped a bit high (or low) depending on how much modification the LUT does to the transfer function relative to the native display transfer function. This results in post-LUT adjustment of brightness usually no more than 1 click up/down on my display.

When I first started using ArgyllCMS for the eeColor box this same issue came up and my first thought was the same as being discussed here, why not embed the user generated (via display controls) transfer function in the source space rec709 icm as your "ideal display". I now agree with Graeme that this should not be done, in other words a profiling/calibration system should be self-contained and target actual standards, not fudged standards.




p.s.
I've also used this technique to modify the transfer function post-calibration.
Edited by zoyd - 10/8/13 at 6:47am
post #805 of 1700
Happy to see improvements are being made to get the perfect workflow for LCDs without Deep Black. Let's hope we soon have our LCD monitors with accurate colors and good contrast. Right now it's not possible it seems to have both.
post #806 of 1700
Thread Starter 
Quote:
Originally Posted by VerGreeneyes View Post


Rec709-1.95.icm - Pure power function with a gamma of 1.95, the same effective gamma as BT.709.
Rec709-2.23.icm - Pure power function with a gamma of 2.23, the same effective gamma as sRGB (basically the same as Rec709_Gamma22.icm).
.

VerGreeneyes,
Using either of these custom icms does produce a 3DLUT with power function but results in bad dE for most color gamut measurements (dE 2-3+). Do you see the same thing when you verify your profile?
post #807 of 1700
Quote:
Originally Posted by N3W813 View Post

VerGreeneyes,
Using either of these custom icms does produce a 3DLUT with power function but results in bad dE for most color gamut measurements (dE 2-3+). Do you see the same thing when you verify your profile?
I haven't tried yet - for the time being I've focused on getting a visually pleasing result, since this has proven so elusive. I'll probably do some measuring during the weekend (for the moment I'm just doing software experiments). FWIW, I use the following commandline:
Code:
collink.exe -v3 -3m -qu -G -Ib:2.2 -iaw -et -Et Rec709-1.95.icm Profile.icm 3DLUT-1.95.icm
or
collink.exe -v3 -3m -qu -G -Ib:2.2 -iaw -et -Et Rec709-2.23.icm Profile.icm 3DLUT-2.23.icm
Note the use of -Ib:2.2 in particular, which seems to work for me but apparently does not for James Freeman (although I think we might be striving for a different ideal). My main uncertainty as far as these settings go is the gamma value to strive for with -Ib - is 2.2 still 'correct' for a profile with a power function of 1.95? Should I be using -Ib:1.95 to match? Should I compensate in the other direction by using 2.2 * (2.2 / 1.95) = 2.48? So yes, I need to spend some time verifying these settings to see what's objectively better.
Edited by VerGreeneyes - 10/8/13 at 10:33am
post #808 of 1700
Quote:
Originally Posted by VerGreeneyes View Post

My main uncertainty as far as these settings go is the gamma value to strive for with -Ib - is 2.2 still 'correct' for a profile with a power function of 1.95? Should I be using -Ib:1.95 to match? Should I compensate in the other direction by using 2.2 * (2.2 / 1.95) = 2.48? So yes, I need to spend some time verifying these settings to see what's objectively better.

Color science theory recommends an end-to-end transfer function between 1.1 and 1.2 assuming "dim surround". This usually means a display transfer function of between 2.2 and 2.4 for BT.709 encoded material. Things get muddled with respect to near black levels but BT.1886 appears to do very well in achieving the 1.1 - 1.2 range for displays with slightly elevated black levels (see for example legacy CRTs with 0.1 cd/m^2 black lift) I'm not a fan of flat display transfer curves even if the display has deep blacks because you'll always end up crushing the first few levels of BT.709 encoded material.
post #809 of 1700
Quote:
Originally Posted by zoyd View Post

I'm not a fan of flat display transfer curves even if the display has deep blacks because you'll always end up crushing the first few levels of BT.709 encoded material.

well...actually....flat are also Sony Trimaster Oled displays... biggrin.gif
post #810 of 1700
Quote:
Originally Posted by zoyd View Post

Graeme has a nice discussion of the alternatives to BT.1886 black point compensation in section 8 and below here, using -ila appearance mapping with or without BT.1886 mapping. You may want to try some of those parameters. Also make sure you have added at least 8 (I use 16) black measurements to your targets to pin down the value in the profile. Even doing that I will run into cases where the LUTed black point will be mapped a bit high (or low) depending on how much modification the LUT does to the transfer function relative to the native display transfer function. This results in post-LUT adjustment of brightness usually no more than 1 click up/down on my display.

When I first started using ArgyllCMS for the eeColor box this same issue came up and my first thought was the same as being discussed here, why not embed the user generated (via display controls) transfer function in the source space rec709 icm as your "ideal display". I now agree with Graeme that this should not be done, in other words a profiling/calibration system should be self-contained and target actual standards, not fudged standards.




p.s.
I've also used this technique to modify the transfer function post-calibration.

Thanks zoyd! Very interesting the fact that you had those same impressions and that finally you've changed your mind, I'll take it as a suggestion.

Actually I ended up in a Excel calculator posted somewhere in a thread. I played with it and I'm realizing that with BT1886 there's a strong relationship between contrast and the rate at which gamma goes up. So displays with very high contrast may show lower black levels (compared to pure power law) even at 10 IRE, or even 5 for best ones.

I will make other test in the following days, especially with my vpr Planar, I suspect that with a good display/projector, BT 1886 becomes very desirable.

Probably it's better getting rid of bad displays rather than BT1886....
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Display Calibration
AVS › AVS Forum › Display Devices › Display Calibration › MadVR - ArgyllCMS