Display Calibration

[Arm3nian]

Continuing from the other thread (http://forum.doom9.org/showthread.php?t=172128&page=23)

As already mentioned, the Spyder 4 can't correctly read black level, or at least reads it in a fashion that is not a standard.

The i1 display pro reads it correctly.

Calibrating with dispcalgui, I set a brightness of 120 cd/m^2, white point of 6500k. End result is somewhere around 6456k, okay, not that big of a deal.

Adjusting RGB sliders around, (most annoying task in the world), I get an average delta E of 0.2. Perfect. After applying the profile, I calibrate again just to see the level. Obviously starting a new calibration temporarily removes and previous profile. With the same monitor settings, now my delta is 0.6... Measuring the delta E with HCFR, my delta E of red for example is 3.2, and average of 2.12. Seriously, wtf?

I don't understand how there is zero consistency between measurements.

The i1 software is fairly bad, and the spyder's is worse. I'm not paying for commercial trash either. So the only thing left is the argyII stuff. Too bad there are basically no guides on anything about it. The dispcalgui wiki doesn't actually tell you what to set, just tells you what each option does. Not very useful if you don't have a Ph.D in the physics of color. All other guides are about something else completely or outdated by 5 years. Same goes HCFR.

What kind of piss poor operation is being run here? The entire color calibration business at this point seems like complete BS to me. Luckily I can still return everything I bought, and I'm fairly close to doing so.

tl;dr: Not impressed by high end displays or calibration techniques. Same result, or worse. Waste of $ and time.

[Asmodian]

Do not target a brightness with the calibration at all, do you have it at ~450cd/m^2 native but are targeting 120 cd/m^2 with the calibration? That would be very bad.

I agree the i1Profiler isn't that great. It is quite important to set the right settings or a calibration can be a lot worse than nothing. I suggest using i1Profiler or dispcal to set your brightness and RGB sliders before calibration using only your monitor's controls.

I use this, it targets the native white point and brightness. Native is optimal for anything but professional color work as long as they are already close to correct.
"C:\Argyllx64\bin\dispcal" -v -dmadvr -qh -ye -G2.4 -f0 -k0 -O "[NAME] matrix" -o "[NAME]_1.icm" "[NAME]"
"C:\Argyllx64\bin\targen" -v -d3 -G -e8 -B8 -s40 -g128 -f1800 -c "[NAME]_1.icm" "[NAME]"
"C:\Argyllx64\bin\dispread" -v -dmadvr -ye -K "[NAME].cal" "[NAME]"
"C:\Argyllx64\bin\colprof" -v -M "[NAME]" -D "[NAME] BT.1886" -qh -aX "[NAME]"
"C:\Argyllx64\bin\collink" -v -M "[NAME]" -D "[NAME] BT.709" -3m -qh -et -Et -IB -G -ims -a "[NAME].cal" Rec709.icm "[NAME].icm" "[NAME] BT.709.icm"

Replace [NAME] with whatever you want but all the [NAME] values should be the same name. It would even work as is but your profiles would have odd names and descriptions. ;)

The Argyll documentation tells you what everything means but please ask if you want to know why I picked any particular setting.

You need to set "-ye" in both the dispcal and the dispread line to your display's backlight type. "e" is white LED on my system.
edit: run "dispcal" without any options to get the list of letters for the backlight types on your system, this is selecting the spectral correction file for your display type.

Also make sure to enlarge the test pattern in the madVR TPG window, it starts very small, both for the dispcal and dispread steps.

The file "Rec709.icm" also needs to be copied into the same directory where you are running these, it is in Argyll's "ref" folder. In the last line, collink, you can use other reference files like "EBU2312_PAL.icm" for EBU/PAL or "SMPTE_RP145_NTSC.icm" for SMPTE C to generate a 3DLUT for other gamuts. I would also change the BT.709 in "[NAME] BT.709.icm" to match. BT.2020 and P3 do not work very well on a nominally sRGB display so I don't think those are useful in your case.

This generates several argyllcms files as well as:

"[NAME]_1.icm", a fast matrix profile - simple windows calibration only.
"[NAME.icm", a profile along with the windows calibration, photoshop and other color managed applications can use this for gamut conversations.
"[NAME] BT.709.icm", a profile that includes the 3DLUT to convert from BT.709 to the monitor. Not useful as far as I know.
"[NAME] BT.709.3dlut, the BT.709 3DLUT for madVR.

[chros]

I don't understand how there is zero consistency between measurements.
I'd be curious about this! If it's the case then it's indeed BS.
(I'm still hesitating whether to buy the 200 pounds value iDisplay Pro or not.)

[huhn]

a meter needs time to heat up and the display too.

so before you calibrated the meter should be an the screen for at least 30 mins and the screen should run at this time too.

i the correct correction loaded?

and most important windows icm profiles are low quality no matter what you do.

[XMonarchY]

For sure - meter does need to be heated up for at least 20-30 minutes and so does the monitor (20-30 minutes). It is also advised to use USB Hub for colorimeters and spectrometers because they need more power than normal USB 2.0/3.0 can provide to work @ their highest potential.

I also recommend dispcalGUI + ArgyllCMS 1.8.2 over using raw ArgyllCMS. DispcalGUI, at this point, is very stable and almost bug-less.

UNINSTALL i1Profiler and delete ALL ICC profiles for now!!! Not only that, but go through your Services and make sure X-Rite services are disabled, along with XRD service (gamma-related). Make sure there is nothing X-Rite related in StartUp either. i1Profiler tries to load ICC profiles upon your PC starting up and that can seriously mess up your calibration.

Use dispcalGUI Install (not stand-alone archive) and once you install and open it up, a menu window should up, asking you if you want dispcalGUI manage ICC profiles. Select "Yes, let dispcalGUI manage profiles".

Are you planning on GAMING on your PC a lot? In that case, there is no need for desktop calibration or ICC profiles. Instead, you should create a 3DLUT for ReShade Framework, which allows the use of such 3DLUT. You can now FULLY calibrate your display for gaming, including both - grayscale AND colorspace / color gamut calibration!

[Arm3nian]

Do not target a brightness with the calibration at all, do you have it at ~450cd/m^2 native but are targeting 120 cd/m^2 with the calibration? That would be very bad.

Well I'm targetting 120 cd/m^2 but I'm also using the dispaly controls to lower the brightness down to it. 450 cd/m^2 is blinding. I tried 140 but it was too bright in my poorly lit room. 120 is a bit low during the day since my display is facing a window. Hard to find a balance but I guess 120 will do.

I agree the i1Profiler isn't that great. It is quite important to set the right settings or a calibration can be a lot worse than nothing. I suggest using i1Profiler or dispcal to set your brightness and RGB sliders before calibration using only your monitor's controls.

Well when the "show advanced options" is unchecked there really isn't that much to configure. One thing I noticed is that dispcal reports my display as 2560x1440 instead of 3840x2160. I have 150% windows scaling on. I don't think that would have any effect though.

I have white level drift compensation on, but have tried it off, and I think on looks better. Black level drift compensation is off. Should I enable this? i1 is really fast so I don't think it slows things down by much. I have spectral: LCD White LED IPS (WLED AC LG Samsung) correction on, which is what my panel is. I target a color temperature of 6500k. The native is 6300k which is a bit off. Isn't one of the main points of calibration to get a correct white point? Tone curve is gamma 2.2 and speed is low. I use auto optimized 2681 patches and high profile quality. Verification is extended chart (51 patches).

When I try it with advanced options I have the black level as measured. Gamma 2.2 relative. Black output offset to 100% and black point correction to 0% and a rate of 4. See anything wrong here?


I use this, it targets the native white point and brightness. Native is optimal for anything but professional color work as long as they are already close to correct.
"C:\Argyllx64\bin\dispcal" -v -dmadvr -qh -ye -G2.4 -f0 -k0 -O "[NAME] matrix" -o "[NAME]_1.icm" "[NAME]"
"C:\Argyllx64\bin\targen" -v -d3 -G -e8 -B8 -s40 -g128 -f1800 -c "[NAME]_1.icm" "[NAME]"
"C:\Argyllx64\bin\dispread" -v -dmadvr -ye -K "[NAME].cal" "[NAME]"
"C:\Argyllx64\bin\colprof" -v -M "[NAME]" -D "[NAME] BT.1886" -qh -aX "[NAME]"
"C:\Argyllx64\bin\collink" -v -M "[NAME]" -D "[NAME] BT.709" -3m -qh -et -Et -IB -G -ims -a "[NAME].cal" Rec709.icm "[NAME].icm" "[NAME] BT.709.icm"

Replace [NAME] with whatever you want but all the [NAME] values should be the same name. It would even work as is but your profiles would have odd names and descriptions. ;)

I'm not calibrating for madVR anymore since I have the new meter. Trying to get a decent system wide calibration before moving on to the 3D LUT. Rec. 1886 gamma 2.4 is what you're supposed to use right? Do you use absolute or relative? Also, when you have madVR selected, the calibration tab has a tone curve option and so does the 3d LUT tab. What should the tone curve in the calibration tab be set to? It is by default "as measured". Is this ignored?

The Argyll documentation tells you what everything means but please ask if you want to know why I picked any particular setting.


What is being changed when you adjust the RGB levels? Is it the white point (color temperature) or are you trying to minimize delta E? Or are both related to each other.

When I lower RGB levels, luminance goes down as expected. So it's kind of annoying to set it to 120cd/m^2, then after adjusting it goes down to 114... or if you have it to 124 and after adjusting it goes to 120. But then I get a delta E of .4 instead of .2.


I'd be curious about this! If it's the case then it's indeed BS.
(I'm still hesitating whether to buy the 200 pounds value iDisplay Pro or not.)
HCFR seems broken to me. Doesn't match spyder, i1 profiler, or dispcalgui. And dispcal is based on argyII just like HCFR.
a meter needs time to heat up and the display too.

so before you calibrated the meter should be an the screen for at least 30 mins and the screen should run at this time too.

i the correct correction loaded?

and most important windows icm profiles are low quality no matter what you do.

For sure - meter does need to be heated up for at least 20-30 minutes and so does the monitor (20-30 minutes). It is also advised to use USB Hub for colorimeters and spectrometers because they need more power than normal USB 2.0/3.0 can provide to work @ their highest potential.

I also recommend dispcalGUI + ArgyllCMS 1.8.2 over using raw ArgyllCMS. DispcalGUI, at this point, is very stable and almost bug-less.

UNINSTALL i1Profiler and delete ALL ICC profiles for now!!! Not only that, but go through your Services and make sure X-Rite services are disabled, along with XRD service (gamma-related). Make sure there is nothing X-Rite related in StartUp either. i1Profiler tries to load ICC profiles upon your PC starting up and that can seriously mess up your calibration.

Use dispcalGUI Install (not stand-alone archive) and once you install and open it up, a menu window should up, asking you if you want dispcalGUI manage ICC profiles. Select "Yes, let dispcalGUI manage profiles".

Are you planning on GAMING on your PC a lot? In that case, there is no need for desktop calibration or ICC profiles. Instead, you should create a 3DLUT for ReShade Framework, which allows the use of such 3DLUT. You can now FULLY calibrate your display for gaming, including both - grayscale AND colorspace / color gamut calibration!

Display is warmed up, so is meter. No USB or other software problems.

When I did make a 3D lut, my movies looked lifeless, like the color was drained. It looked like my 10 year old piece of crap TN panel. It was not at all enjoyable. Contrast ratio affects colors other than black but it shouldn't look this bad at 1050:1. Something is definitely wrong. I'll try again with the 3DLUT because it is easier to tell if the calibration succeeded. Testing windows profile is harder.

[Arm3nian]

I profiled and made a 3D lut following the AVS guide. I see no difference. Still looks like crap. Games look fine to me, without any calibration.

The low bitrate stream from cable going into the living room TV looks better than what I'm watching. This is really depressing.

[Asmodian]

Did you try my settings? The batch file I posted is for doing a normal calibration, only the last line is for creating the 3DLUT for madVR. Feel free to increase the patch count to 2681 if you want to (change the -f1800 to -f2681 in the targen line). I don't like the newer dispcalgui versions because I don't know what it is doing.

On my system HCFR matches perfectly with Argyllcms, they also use the same meter reading code. Are you using madTPG and the same spectral correction everywhere?

Changing the RGB sliders is to change the white point, which also changes dE measurements. Are you not able to get close to 6500K by only adjusting the RGB sliders? You do have to adjust brightness and the RGB sliders together, that is simply how it is. Any dE below 1 is quite good, simply moving the meter on the screen will often change a 0.2 dE to 0.4 or the other way around. Most displays aren't very uniform. :(

Don't use white or black point correction, that is mostly for Plasma, or at least not a warmed up LCD.

[Arm3nian]

Those settings are what I used when I tried again. It looks the same as before.

I'm going to return my display. I don't see a point in 4k anymore. Nothing scales correctly when windows is at 150% and if it isn't then you can't see anything on a 27inch. Originally I thought non integer scaling looks bad, but with madVR, it looks good. 1440->4k looks great with doubling. 1080->1440 should also look good. 10bit also isn't very useful since most of the time I'm doing things other than watching movies. I don't notice much banding on a true 8bit display.

Maybe I'll get the new ROG swift (pg279q). I'm convinced at this point that almost all IPS look the same. No point in paying $700 for my current display when I can get the same thing but gsync/ulmb and 165Hz refresh rate for almost the same price. I'll also probably return the i1. It looks the same as the spyder even though it can read black. Maybe I'll callibrate my new display before sending it back. :p

[Arm3nian]

Why didn't I think of this before... too busy arguing with smart asses on oc.net. Here are screenshots of no calibration vs 3D LUT.

Pics from the hobbit (imgur compression kind of killed quality but the color differences are still visible)
Group 1: normal (http://i.imgur.com/0h2HAJ8.jpg) vs 3D LUT (http://i.imgur.com/ONL1ph6.jpg)
Group 2: normal (http://i.imgur.com/MxC8THH.jpg) vs 3D LUT (http://i.imgur.com/pqYJ158.jpg)

The ones without the LUT look darker, I guess you could say saturated. Calibrated results look more natural? Seems a bit dull though. I tried some other sources and the same results.

[Asmodian]

I don't think you could have used my command lines to generate those images. Especially if you want an image more like un-calibrated you really should try them.

However, calibration can only remove total saturation. For example, the way you fix the white point with a calibration is by reducing the saturation of the two colors that are too saturated. This is why you don't target a white point unless it is off enough to bother you. Human vision adapts white anyway so minor errors are not a big deal, the loss of contrast looks worse. In the mid-range it is possible to add saturation and a good calibration can fix a bad gamma which can make the image seem higher contrast.

If you want a bit deeper/darker shadows try this:
"C:\Argyllx64\bin\dispcal" -v -dmadvr -qh -ye -G2.4 -f0.6 -k0 -O "PG278Q matrix" -o PG278Q_1.icm PG278Q
"C:\Argyllx64\bin\targen" -v -d3 -G -e8 -B8 -s40 -g128 -f2259 -c PG278Q_1.icm PG278Q
"C:\Argyllx64\bin\dispread" -v -dmadvr -ye -K PG278Q.cal PG278Q
"C:\Argyllx64\bin\colprof" -v -M "ROG Swift PG278Q" -D "PG278Q BT.1886" -qh -aX PG278Q
"C:\Argyllx64\bin\collink" -v -M "ROG Swift PG278Q" -D "PG278Q BT.709" -3m -qh -et -Et -IB:0.6:2.4 -G -ims -a PG278Q.cal Rec709.icm PG278Q.icm "PG278Q BT.709.icm"

[Arm3nian]

So targeting specific settings is only if you need it? The screenshots were with a targeted white point of 6500k, not the default, which is more around 6356k. I'll do it again when I have time.

[Asmodian]

So targeting specific settings is only if you need it? The screenshots were with a targeted white point of 6500k, not the default, which is more around 6356k. I'll do it again when I have time.

Monitor calibration methodology has been designed for matching different displays to each other and to print work, not to give the best viewing experience. What makes a good calibration for matching prints is not the same as what provides an optimal experience watching video.

I, and others who's screens I have calibrated, much prefer the calibration when not targeting a white point or (shudder) a brightness. Targeting a brightness with the calibration for viewing video is simply a bad idea. There is no way you can notice a small change in brightness but you reduce saturation on all colors in order to target a brightness, which reduces contrast too. The only way to reduce brightness with a calibration is to have the video card output 254 or less when the source pixel's value is 255, this requires lowering all values to avoid banding, reducing the dynamic range for no visible benefit.

White point calibration is more debatable. My view is that you should only target a white point if the current white has a color tint you do not like, and then only maybe. For almost all my calibrations on displays with a reasonable white point (~6200 to 6800K), targeting a white point does not improve my perception of white but it does make the image look more faded or washed out, like in your screen shots. Tuning with the RGB sliders on the display, with feedback from a meter, has always allowed me to get a good enough white point so I do not need to target one with the calibration. The sad part is that on displays with a really bad white point (e.g. >9000K) targeting a white point reduces contrast and saturation enough that most people have preferred not targeting a white point even in extreme cases. If blue can only use 0-217 and green 0-188 it reduces the saturation of all colors by a lot.

I have enough experience trying to notice display color accuracy that I do get bothered by white points too far from 6500K and I do prefer them calibrated to 6500K, even with the loss of contrast and saturation. I constantly notice the bad white instead of watching the video. :o

I have handled it by giving away or returning all my displays with bad white points, rather than calibrating them into a faded mess. :p

This may give the impression that I think calibrations are somewhat useless but the calibrated gamma curve is generally much better, as long as you do not get banding, and tuning the white point with the meter and the monitor's controls often helps a lot.

[Arm3nian]

Now I understand. So "as measured" simply uses what you set the white point and white level to be via monitor controls. And targeting a value means adjustments are made in software to get you where you couldn't with the monitor controls.
I thought that "as measured" meant it used the initial reading without adjustments and targeting meant it just brought up a value and told you if you were close or not... fail :o
I can get very close to 6500k and 120cd/m^2 with a delta E of 0.2-0.3 using the sliders.

I'm guessing as measured is recommended for system wide calibration as well? What should I use for the tone curve for a windows profile, gamma 2.2 or as measured (I set the settings to default (gamma 2.2) in the top)? I am not doing pro image work or printing, just want nice colors. I also don't use the advanced options as they seem to make things worse if you don't need it. Is auto-optimized 2681 patches good for windows?

[Asmodian]

Now I understand. So "as measured" simply uses what you set the white point and white level to be via monitor controls. And targeting a value means adjustments are made in software to get you where you couldn't with the monitor controls.
I thought that "as measured" meant it used the initial reading without adjustments and targeting meant it just brought up a value and told you if you were close or not... fail :o
I can get very close to 6500k and 120cd/m^2 with a delta E of 0.2-0.3 using the sliders.

I'm guessing as measured is recommended for system wide calibration as well? What should I use for the tone curve for a windows profile, gamma 2.2 or as measured (I set the settings to default (gamma 2.2) in the top)? I am not doing pro image work or printing, just want nice colors. I also don't use the advanced options as they seem to make things worse if you don't need it. Is auto-optimized 2681 patches good for windows?

Correct, what I do is start a calibration with a target white point, setup the monitor, and then cancel and restart using native/as measured.

Also definitely do the same for the normal system calibration, I would argue it is even more important to not limit saturation globally. For gamma I think 2.4 actual (-G2.4) or 2.2 effective (-g2.2) are both reasonable targets for a Windows calibration. I prefer actual but effective is better for matching displays with different black levels.

2681 patches is probably overkill to be honest, but the only reason to use fewer is if you get banding with too many (speed too, of course). :)

[markanini]

I curious if anyone here uses f.lux together with a display profile and/or MadVR 3DLUT. Sometimes I turn it off for color critical work but most of the time it's on and does a excellent job and reducing eye strain. So I've been concerned with what kind of implication this might have on image quality. The f.lux filtering seems to sit on top of both the 3dlut and display profile so I guess it's an overlay, which should allow a luma corrected filter. But how precise can it be when it's not aware of the specific display characteristics?

[Asmodian]

I curious if anyone here uses f.lux together with a display profile and/or MadVR 3DLUT. Sometimes I turn it off for color critical work but most of the time it's on and does a excellent job and reducing eye strain. So I've been concerned with what kind of implication this might have on image quality. The f.lux filtering seems to sit on top of both the 3dlut and display profile so I guess it's an overlay, which should allow a luma corrected filter. But how precise can it be when it's not aware of the specific display characteristics?

If you are using f.lux do not bother with any kind of calibration. f.lux is only for those who do not care about color accuracy; it completely destroys the dE values of any calibration.
edit: Actually f.lux on top of a calibration is probably better than without the calibration; f.lux does try to do a good job visually, even if it cannot be accurate, so it is still likely to look better after the calibration. However, it is especially important to not target a white point with the software calibration when using f.lux.

This is expected based on how it works; it is doing its own calibration without any understanding of the display, as you mentioned, and it is targeting reduced eye strain instead of color accuracy. It wants color accuracy to be bad, you cannot arbitrarily remove blue light and get accurate colors.

There have also been reports of f.lux interacting badly with madVR, causing dropped frames and/or choppiness.

I would do a normal calibration but targeting 5003K (illuminant D50, x0.34567, y0.35850) or maybe 5503K (D55, x0.33242, y0.34743) for the white point if I wanted a low blue light calibration. This might not be as low eye-strain as using f.lux on top of a calibration targeting 6503K (D65, x0.31271, y0.32902) but it would give accurate colors relative to white.

[Arm3nian]

I calibrated again using "as measured", but using targeted values and then cancelling before to get my desired levels. After setting levels, my color temperature was 6484K, white level was 119cd/m^2, and delta E was .1 to .3. After, measuring with HCFR, my color temperature went lower to 6450K, white level went up to 121cd/m^2, and delta E of white is 1. So not bad, but why is it worse than before calibrating? Contrast ratio is around 1126:1.

The individual delta E's are 3.4, 3.8 and 1.1 for RGB. Why are they so high? Do you get the same results?

Turning the profile on or off makes no changes in HCFR readings...
To compare subjectively, the calibration looks the same as before except a bit dimmer. It looks good, but the objective measurements are fishy.

Speaking about eye care, my display has an internal blue light filter that is set to 100 by default. How does this affect color accuracy?

[huhn]

do you use a correction and is HFCR using the SAME correction?

[Asmodian]

I calibrated again using "as measured", but using targeted values and then cancelling before to get my desired levels. After setting levels, my color temperature was 6484K, white level was 119cd/m^2, and delta E was .1 to .3. After, measuring with HCFR, my color temperature went lower to 6450K, white level went up to 121cd/m^2, and delta E of white is 1. So not bad, but why is it worse than before calibrating? Contrast ratio is around 1126:1.

In addition to what huhn asked, are you sure the meter is in exactly the same spot and the display has warmed up for a similar amount of time? Changes on that order are common when moving the meter, in my experience.

The individual delta E's are 3.4, 3.8 and 1.1 for RGB. Why are they so high? Do you get the same results?

This is perfectly normal, 100% Red, Green, and Blue are never correct on normal consumer displays. Apparently it is very hard to make white LEDs with exact primaries. Actually those are pretty good, I often see almost 5 dE on the worst primary. My current displays (both have pretty new white LED backlights) have basically the same error in the primaries you do.

Turning the profile on or off makes no changes in HCFR readings...
To compare subjectively, the calibration looks the same as before except a bit dimmer. It looks good, but the objective measurements are fishy.

Are you using madTPG with HCFR as well? Make sure you don't have it set to disable calibrations. Also make sure you are using the same spectral correction and don't move the meter between calibration and measuring with HCFR. Look at the gamma, that should have the biggest change when using native targets. Usually a native white point calibration is not an obvious change unless the native gamma is very bad.

Speaking about eye care, my display has an internal blue light filter that is set to 100 by default. How does this affect color accuracy?

I don't know, assuming it was on when you calibrated it sounds like it is fine, a max dE of 3.8 isn't bad. What is the dE for the cyan and magenta saturation measurements? What happens to the appearance and HCFR measurements with it off?

[chros]

I need your help about a used OEM i1 Display Pro device. What does it mean if it has a missing diffuser head? Can it be still used for calibrating?
Thanks

[foxyshadis]

I need your help about a used OEM i1 Display Pro device. What does it mean if it has a missing diffuser head? Can it be still used for calibrating?
Thanks

You should contact the company about errors like that: http://www.xrite.com/i1display-pro/Support

[Asmodian]

I need your help about a used OEM i1 Display Pro device. What does it mean if it has a missing diffuser head? Can it be still used for calibrating?
Thanks

I am not sure about the "head" part of diffuser head? If it is only the ambient light diffuser having it missing simply means you need to be more careful with the meter when not calibrating because it acts like a cover for the lens when not in use.

The diffuser is useful when calibrating against prints as you need to adjust for ambient lighting (prints look different in different light so your screen should too). However, for normal video watching, I have never liked the results when adjusting for ambient lighting.

I never use the diffuser when calibrating normal displays, only as a lens cover when I am done.

[markanini]

What about the difference between sRGB recommended chromaticity coordinates (0.3127x0.3290) and a straight 6500k?

[Asmodian]

What about the difference between sRGB recommended chromaticity coordinates (0.3127x0.3290) and a straight 6500k?

No one can see it. :)

Illuminant D65 is (0.31271x0.32902) so BT.709/sRGB is simply the values rounded to four decimal places. D65 is actually 6503.6K.

[chros]

The diffuser is useful when calibrating against prints as you need to adjust for ambient lighting (prints look different in different light so your screen should too). However, for normal video watching, I have never liked the results when adjusting for ambient lighting.

I never use the diffuser when calibrating normal displays, only as a lens cover when I am done.
Good news, thanks for the info! I'll get it one of the days and I'll play with it. (I also bought an i1pro , it'll take more time to get here.)
And I'm reading a lot in the meantime, though I feel more confused than I was at the beginning (but probably it's a good thing :) )

1. I want to calibrate my TV, it has 2 points/10 points IRE controls for grayscale calibration and a HLS CMS system for colors. Can I do this with Argyll/DispalGUI or HFCR as the 1st step (calibration)? I know you can set RGB values in DispalGUI but can you do such an advanced calibration as well? (I went through the grayscale/color calibration, but it uses ChromaPure for this: http://www.curtpalme.com/forum/viewtopic.php?t=35322 )
2. In the above guide was said that it's a good practice to put the meter on a tripod even if you don't have projector. What do you think? (I'm asking because the CCFL S-IPS panel can heat up well in time.)
3. My TV crushes black, so I have to set custom levels in MadVR like this: black: 1,255 . (It's even worse with the laptop builtin LED blacklight panel, I have to set black to: 5,255 .) huhn said in the avs forum topic, that we should leave it in this way before calibration. Is it a good practice? (Since that would mean that we'll get inaccurate colors for the icm profile.)

Sorry for the silly questions, there'll be more later probably :)
Thanks

[markanini]

What's the purpose of the tripod? Just for eliminating ambient light from getting into the sensor? You could try it both ways regarding the black crush and see which gives you the lowest deltaE or the least banding on a gradients.

[huhn]

3. My TV crushes black, so I have to set custom levels in MadVR like this: black: 1,255 . (It's even worse with the laptop builtin LED blacklight panel, I have to set black to: 5,255 .) huhn said in the avs forum topic, that we should leave it in this way before calibration. Is it a good practice? (Since that would mean that we'll get inaccurate colors for the icm profile.)

only use this for 3D LUT calibration.

[chros]

So, first of all, thank you all for your suggestions. After fiddling around about 3 days with the i1DisplayPro (a new one, the i1Pro is still on its way), finally I reached the point where I can raise questions. :)
I mostly measured during evening/night so I settled down with just setting brightness(53), contrast(89), backlight(30), color gamut(bt.709), color temp (warm) and gamma (2.4) for now and leaving everything else on default (after resetting an ISF mode) (that means I didn't set any grayscale/CMS setting) and I left them at these settings until I'll be comfortable with the rest. I went for Asmodian's suggestion about leaving white point as it is (~6240K) and the resulted brightness is around 106 cd/m2. So far so good. I used the standard CCFL-IPS correction everywhere. I used blacklevel, apl-clipping, white-level, color-clipping mp4 videos from AVSHD to check these original values and to check the result at the end of every calibration/profiling.
The goal would be to get a good looking picture for now with the above setting and with 3dlut.
About the black crush of mine: I made all the measurements with the slightly raised black level (custom level: 1-255 in madvr) for the sake of simplicity.

1.
When did make a 3D lut, my movies looked lifeless, like the color was drained. It looked like my 10 year old piece of crap TN panel.
I figured out what is the problem with the default madvr preset of dispalGUI. I got the same result as you did, no matter what I did.
On the 3Dlut 'tab' at Rendering intent 'Absolute Colorimetric with white point scaling' is selected by default (this is '-iaw' switch for collink.exe) instead of Asmodian's '-ims' switch which would be the 'Saturation' settings in dispalGUI. (Argyll docs http://www.argyllcms.com/doc/collink.html#i)
- I have tried out couple of Absolute like intents (-iaw , -ia , -ir) and they behave the same way, resulting slight fog on top of the picture
- I tried out couple of other intents from the other group (-ims , -ip , -is) and, they behave differently, resulting NO fog at all
You can run manually the last step to fix your 3dlut by modifying the collink.exe command line from DispalGUI.

2.
Do not target a brightness with the calibration at all, do you have it at ~450cd/m^2 native but are targeting 120 cd/m^2 with the calibration? That would be very bad.
...
I use this, it targets the native white point and brightness. Native is optimal for anything but professional color work as long as they are already close to correct.
"C:\Argyllx64\bin\dispcal" -v -dmadvr -qh -ye -G2.4 -f0 -k0 -O "[NAME] matrix" -o "[NAME]_1.icm" "[NAME]"
"C:\Argyllx64\bin\targen" -v -d3 -G -e8 -B8 -s40 -g128 -f1800 -c "[NAME]_1.icm" "[NAME]"
"C:\Argyllx64\bin\dispread" -v -dmadvr -ye -K "[NAME].cal" "[NAME]"
"C:\Argyllx64\bin\colprof" -v -M "[NAME]" -D "[NAME] BT.1886" -qh -aX "[NAME]"
"C:\Argyllx64\bin\collink" -v -M "[NAME]" -D "[NAME] BT.709" -3m -qh -et -Et -IB -G -ims -a "[NAME].cal" Rec709.icm "[NAME].icm" "[NAME] BT.709.icm"
Thanks for your batch script! It help me a lot.
But unfortunately I got 2 problems with it (first has been fixed):
a) I got really nasty black levels: black-level (I could see 14-16 (!) bars) and white-clipping (the "nonvisible" part of the blacks were also flashing!) test was really bad with this! Log of DispalGUI helped me with this, if I use the "-b" switch of dispcal.exe (Use forced black point hack , http://www.argyllcms.com/doc/dispcal.html#bhack) in the first command then these nasty black levels are vanished! (DispalGUI also use this '-b' switch for the last command (collink.exe) but I coudn't see any difference with/without it.)
b) when looking at the black-level test (with the above modification), 17-19 bars are gone! All the rest of the bars looks natural and all the other test clips are looking good as well. Can it be that Francois76l was talking about the same thing? http://forum.doom9.org/showthread.php?p=1747776#post1747776
I tried out the '-ims -IB:0.1:2.4' combination for "collink.exe" but there was no difference.
I have only this problem with your method, and since I originally using the raised balck levels in madvr it's clearly not good that the calibrated picture lacks of blacks. Any idea?

c) Can we verify somehow the result (like we can with dispalGUI method)?

3.
only use this for 3D LUT calibration.
I don't really know what this means. What part of the process do you mean?
a) with the above command line: do you mean the 3rd command (dispread.exe)? (there's another reading during the dispcal.exe command)
b) with dispalGUI the situation is pretty interesting, it uses different method: setting measured (as target) everywhere (the full log is here: http://pastebin.com/nPrEspS1 )
here are the commands (timecodes are included):
00:38:44,208 dispcal.exe -F "-P0.5,0.5,9.486833" -v2 -dmadvr -c1 -yn -X CCFLFamily_07Feb11.ccss -t -b "madVR 2015-12-06 00-38 S XYZLUT"
00:39:30,080 dispread.exe -F "-P0.5,0.5,9.486833" -v -dmadvr -c1 -yn -X CCFLFamily_07Feb11.ccss -K "madVR 2015-12-06 00-38 S XYZLUT.cal" "madVR 2015-12-06 00-38 S XYZLUT"
00:42:46,055 colprof.exe -v -qm -ax -bn "madVR 2015-12-06 00-38 S XYZLUT"

00:42:59,679 targen.exe -v -d3 -e4 -s33 -g97 -m0 -f1399 -A1.0 -G -N0.5 -B4 -b0 -V1.6 -c "madVR 2015-12-06 00-38 S XYZLUT.icm" "madVR 2015-12-06 00-38 S XYZLUT"
00:43:50,631 dispread.exe -F "-P0.5,0.5,9.486833" -v -dmadvr -c1 -yn -X CCFLFamily_07Feb11.ccss -K "madVR 2015-12-06 00-38 S XYZLUT.cal" "madVR 2015-12-06 00-38 S XYZLUT"
01:10:04,398 colprof.exe -v -qh -ax -bl -C "No copyright. Created with dispcalGUI 3.0.5.0 and Argyll CMS 1.8.3" -M madVR -D "madVR 2015-12-06 00-38 S XYZLUT" "madVR 2015-12-06 00-38 S XYZLUT"

01:11:12,024 iccgamut.exe -v -w -ir -ff -on -d6.00 "madVR 2015-12-06 00-38 S XYZLUT.icm"
01:11:15,547 viewgam.exe -cw -t0 -w "C:\Users\chros\AppData\Local\0install.net\implementations\sha1new=1a6691a40
4d21b03d010dc8de1fe023e81133617\dispcalGUI\ref\sRGB.gam" -cn -t.3 -s "madVR 2015-12-06 00-38 S XYZLUT.gam" -i "madVR 2015-12-06 00-38 S XYZLUT vs sRGB.wrl"
01:11:21,889 viewgam.exe -cw -t0 -w "C:\Users\chros\AppData\Local\0install.net\implementations\sha1new=1a6691a40
4d21b03d010dc8de1fe023e81133617\dispcalGUI\ref\ClayRGB1998.gam" -cn -t.3 -s "madVR 2015-12-06 00-38 S XYZLUT.gam" -i "madVR 2015-12-06 00-38 S XYZLUT vs ClayRGB1998.wrl"
01:11:29,981 collink.exe -v -qh -G -iaw -r65 -n -3m -et -Et -b -IB:0.0:2.4 -a "madVR 2015-12-06 00-38 S XYZLUT.cal" Rec709.icm "madVR 2015-12-06 00-38 S XYZLUT.icm" "madVR 2015-12-06 00-38 S XYZLUT.Rec709.B0.0,2.4Gawt65.icm"
(3dlut is installed into madVR)

The interesting part is the first 3 command: it uses dispcal only for getting the "measured settings" then it quits, then it uses dispread and colprof for the rest (what dispcal did in the first command of the command line version.)
c) So at which point would I suppose to switch the custom levels on?

Otherwise, setting Saturation for Rendering intent in dispalGUI, resulted a good looking picture, but it had color clipping in the blue and suffered the same missing 17-19 black bars as the command line version (so this result was worse :) ).

4. About ccss correction files (until the i1pro will be arrived):
My TV is supposed to be a wide gamut CCFL 8+dither(or FRC) IPS. So selecting the correct correction file isn't an easy task to do either.
a) dispalGUI reports for white point (no matter whether I set 'bt.709' or 'wide' for color gamut on TV device):
- with CCFL-IPS: correct RGB balance (~dE 0.2) and ~6240 K
- with WGCCFL-IPS: wrong RGB balance and ~6100 K
This difference is huge, especially you want to correct the RGB Contrast with TV controls.
b) in HCFR checking the color gamut diagrams:
- with CCFL-IPS AND 'bt.709' (in TV) : resulting gamut is smaller than sRGB
- with CCFL-IPS AND 'wide' (in TV) : resulting gamut is bigger than sRGB
- with WGCCFL-IPS AND 'bt.709' (in TV) : resulting gamut is smaller than sRGB (that's understandable)
- with WGCCFL-IPS AND 'wide' (in TV) : resulting gamut is about the same as sRGB, slightly rotated

That makes me think how we can trust the supplied ccss files. The same colors are presented on the display (per color gamut settings) but you'll get different results, so you'll adjust colors completely differently on the very same display!

c) How can we create ccss file with command line?
What "default" setting should I use with the TV to do it?

Thanks for you help and for reading :)

[huhn]

with custom ranges you get an incorrect ICM that's pretty much useless for everything else than madVR.

when you use custom levels you have to use it from the start.

[chros]

with custom ranges you get an incorrect ICM that's pretty much useless for everything else than madVR.
I figured :) But it's not a problem since I use it only on the TV that is used just by madVR.
when you use custom levels you have to use it from the start.
OK, thanks, that's what I did.

[Siso]

So in general what do you guys recommend to start a dispcal guimonitor calibration and after that a madvr 3dlut? Seems like there are a lot of different opinions...Any advices would be appreciated :)

And what about these white level drift compensation and black level drift compensation? And shouldn't the curve for the monitor gamma 2.2 should be srgb curve? I'm new to the calibration :D gonna calibrate my Dell U2913WM thingie :D

BTW is this a good option for madvr 3dlut - https://imouto.my/tutorials/madvr/#3dlut ?

[huhn]

And what about these white level drift compensation and black level drift compensation?

this shouldn't be needed for an LCD monitor.

windows assumes sRGB no question in realty it isn't really used. it depends on what you want to display.

in the end gamma 2.2 isn't really used too. it's gamma 2.4 and BT 1886 for complicated reasons.

BTW is this a good option for madvr 3dlut - https://imouto.my/tutorials/madvr/#3dlut ?

it doesn't look like there is anything wrong with it but i wouldn't use it anyway.

i personally use 5000-10000 patches

this is a nice thread: http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms.html

calibration with dispcal GUI is pretty easy.

1. find the correct correction for your display type.

2. fix the white point manually with the "interactive display adjustment".

3. let dispcal GUI do the rest.

[Siso]

this shouldn't be needed for an LCD monitor.

windows assumes sRGB no question in realty it isn't really used. it depends on what you want to display.

in the end gamma 2.2 isn't really used too. it's gamma 2.4 and BT 1886 for complicated reasons.



it doesn't look like there is anything wrong with it but i wouldn't use it anyway.

i personally use 5000-10000 patches

this is a nice thread: http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms.html

calibration with dispcal GUI is pretty easy.

1. find the correct correction for your display type.

2. fix the white point manually with the "interactive display adjustment".

3. let dispcal GUI do the rest.

Sounds good, for testcharts for the windows calibration - large or extra large? Any special options for curves+matrix or I should use xyz lut+matrix?

[huhn]

i don't know i gave up on windows calibration. i just click on sRGB and use the defaults.

[Siso]

i don't know i gave up on windows calibration. i just click on sRGB and use the defaults.

I see, thank you for the answers :)

[chros]

calibration with dispcal GUI is pretty easy.
1. find the correct correction for your display type.
2. fix the white point manually with the "interactive display adjustment".
3. let dispcal GUI do the rest.
In addition to what huhn said, this is the official guide:
http://sourceforge.net/p/dispcalgui/wiki/3D%20LUT%20creation%20workflow%20for%20madVR%20or%20eeColor/
Try it out at first :) (I don't know the answer to your other questions.)
What kind of instrument(s) do you have?

And I have posted new result in here, if anybody is interested: http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms-137.html#post39630546
I'm getting closer but still not enough :)

[Asmodian]

a) I got really nasty black levels: black-level (I could see 14-16 (!) bars) and white-clipping (the "nonvisible" part of the blacks were also flashing!) test was really bad with this! Log of DispalGUI helped me with this, if I use the "-b" switch of dispcal.exe (Use forced black point hack , http://www.argyllcms.com/doc/dispcal.html#bhack) in the first command then these nasty black levels are vanished! (DispalGUI also use this '-b' switch for the last command (collink.exe) but I coudn't see any difference with/without it.)

This is normal, black level is a complicated issue as you found. I used to have to use -b as well but with my new displays and/or new versions of Argyllcms I do not.

b) when looking at the black-level test (with the above modification), 17-19 bars are gone! All the rest of the bars looks natural and all the other test clips are looking good as well. Can it be that Francois76l was talking about the same thing? http://forum.doom9.org/showthread.php?p=1747776#post1747776
I tried out the '-ims -IB:0.1:2.4' combination for "collink.exe" but there was no difference.
I have only this problem with your method, and since I originally using the raised balck levels in madvr it's clearly not good that the calibrated picture lacks of blacks. Any idea?

If you need the black point hack you almost always get crushed shadows. You probably simply have to live with slightly crushed blacks as your display cannot get both an accurate gamma and desirable behavior near black. To balance this I use something like -IB:0.5:2.4, 0.1 isn't significant enough to make a visible difference. -IB:1.0:2.4 would be all crushed while -IB:0.0:2.4 (equivalent to -IB) would be raised shadow levels (gray shadows). Use -f0.5 in dispcal if using -IB:0.5:2.4 in collink.
edit: I should have said use -f0.5 in dispcal to match -IB:0.5:2.4 in collink, it is reasonable to use different values for the calibration and 3DLUT, but possibly not optimal.

Even with -f0 in dispcal you shouldn't get a raised black point when using -k0 and not targeting a white point. If you want to target a white point and get a raised black point even with -k0 use -A set to something below 4.0 and above 1.0, e.g. -A2.5.

c) Can we verify somehow the result (like we can with dispalGUI method)?

I use HCFR to measure my displays after calibration.

Sorry for the late response. :o

[markanini]

I wonder what implications -IB:x.x:2.4 switch or output offset slider has for stuff level adjustments etc. I assume defaults are the most accurate but I too have found the shadows to be somewhat lifted on my IPS panels so I'm tempted to try reducing that now.

[Asmodian]

The -IBx.x:2.4 value has a fairly subtle overall effect on ~0.15 cd/m² black level (~1000:1) or lower displays. It is noticeable but only in shadows.

Even using -IB (equivalent to -IB0.0:2.4) you should still see no bar for 16 and below, and 17 and higher should be easy to see, 17 might be too easy to see (gray shadows). As you raise x.x you push 17+ closer to 16, depending on the display's native behavior you can even crush 18-19 into 16 as you raise it to 1.0 (equivalent to -IG:2.4).

If you can see 16 or below something is wrong with your calibration because it raised the black point. Make sure to use -k0 for dispcal and lower -A4.0 if you have to. If you need to use -b (forced black point hack) you will probably have crushed shadows and be unable to see 17 even with -f0/-IB. Personally, losing 17-18 is optimal on some displays. Some shadow detail is lost but everything looks normal and the apparent contrast is much better; avoid raising the black point at all costs.

[Siso]

In addition to what huhn said, this is the official guide:
http://sourceforge.net/p/dispcalgui/wiki/3D%20LUT%20creation%20workflow%20for%20madVR%20or%20eeColor/
Try it out at first :) (I don't know the answer to your other questions.)
What kind of instrument(s) do you have?

And I have posted new result in here, if anybody is interested: http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms-137.html#post39630546
I'm getting closer but still not enough :)

I'm trying to create a 3dlut for madvr, your results look OK, can you share more info on your attempt to create a good 3dlut for madvr?

Here are some settings from my latest windows calibration:

http://s21.postimg.org/ghygjrvcj/le_1.png (http://postimg.org/image/ghygjrvcj/)

http://s21.postimg.org/bgr2ienw3/le_2.png (http://postimg.org/image/bgr2ienw3/)

http://s21.postimg.org/afqtta6wj/le_3.png (http://postimg.org/image/afqtta6wj/)

[chros]

If you need the black point hack you almost always get crushed shadows.

I wonder what implications -IB:x.x:2.4 switch or output offset slider has for stuff level adjustments etc.

The -IBx.x:2.4 value has a fairly subtle overall effect on ~0.15 cd/m² black level (~1000:1) or lower displays. It is noticeable but only in shadows.
Thanks for the detailed explanation, Asmodian!
Well, the -IB:x.x:2.4 switch depends on which rendering intent do you use! (-i switch) Asmodian used -ims , dispalGUI offers -iaw by default. The difference is the -ims is relative, while -iaw is absolute. They state that -iaw supposed to be more accurate (I didn't had the time to test it on my setup though). But, -IB:x.x:2.4 works the opposite way for these 2 groups: to eliminate the fog effect of -iaw I had to set the slider to 100 (that means IB:1.0:2.4).
About the black crush: I noticed that the cause of it is the "-G2.4 -f0 -k0" switches in the dispcal command. If I leave it out and '-b' switch is also present I won't get black crush.


I'm trying to create a 3dlut for madvr, your results look OK, can you share more info on your attempt to create a good 3dlut for madvr?
In dipalGUI, select your correction matrix then follow the tutorial on link link you've posted. https://imouto.my/tutorials/madvr/#3dlut
So far this gives me the best result (with the custom testchart) with 2 modifications on my old CCFL-IPS display:
- on calibration tab: leave Tone curve "as measured"
- on 3dlut tab: set 100% for the black offset (under rendering intent) (you have to enable the advanced settings in the menu, I think)
With this I don't get the black crush (17-19) and there isn't any fog-effect, and thanks to the custom testchart I got the most accurate colors (so far, at least :) ).

[Siso]

Thanks for the detailed explanation, Asmodian!
Well, the -IB:x.x:2.4 switch depends on which rendering intent do you use! (-i switch) Asmodian used -ims , dispalGUI offers -iaw by default. The difference is the -ims is relative, while -iaw is absolute. They state that -iaw supposed to be more accurate (I didn't had the time to test it on my setup though). But, -IB:x.x:2.4 works the opposite way for these 2 groups: to eliminate the fog effect of -iaw I had to set the slider to 100 (that means IB:1.0:2.4).
About the black crush: I noticed that the cause of it is the "-G2.4 -f0 -k0" switches in the dispcal command. If I leave it out and '-b' switch is also present I won't get black crush.

In dipalGUI, select your correction matrix then follow the tutorial on link link you've posted. https://imouto.my/tutorials/madvr/#3dlut
So far this gives me the best result (with the custom testchart) with 2 modifications on my old CCFL-IPS display:
- on calibration tab: leave Tone curve "as measured"
- on 3dlut tab: set 100% for the black offset (under rendering intent) (you have to enable the advanced settings in the menu, I think)
With this I don't get the black crush (17-19) and there isn't any fog-effect, and thanks to the custom testchart I got the most accurate colors (so far, at least :) ).

Will try that these days, and I'll share my results :), thank you.

A side note about the windows calibration, is it OK (the pics I had uploaded)?

[chros]

A side note about the windows calibration, is it OK (the pics I had uploaded)?
Which pic exactly? I was talking about them. :)

[Siso]

Which pic exactly? I was talking about them. :)

Oh, sorry then :D Will make the 3dlut tonight to see how the things are :)

[Siso]

Here are the results from tonight:

http://s28.postimg.org/auu40fe21/with.png (http://postimg.org/image/auu40fe21/)

http://s28.postimg.org/i0lv2vn55/without.png (http://postimg.org/image/i0lv2vn55/)

black clipping test:

http://s9.postimg.org/s4plxtkfv/black_clipping_3dlut_off.png (http://postimg.org/image/s4plxtkfv/)

http://s9.postimg.org/afxv678or/black_clipping_3dlut_on.png (http://postimg.org/image/afxv678or/)

[Asmodian]

Screen shots don't show us anything, they look different on my screen than they do on your's. ;)

[Siso]

Screen shots don't show us anything, they look different on my screen than they do on your's. ;)

I guess it depends on the calibration, I thought one of the purposes to calibrate your screen is to close match other people's calibrated screens :D anyway, I don't have visible bars from 17-19 with the 3dlut, without it, they are visible.

[huhn]

a 3d lut creates an image that looks right on your screen and only on your screen.

[raffriff42]

Monitor calibration methodology has been designed for matching different displays to each other and to print work, not to give the best viewing experience. What makes a good calibration for matching prints is not the same as what provides an optimal experience watching video.

I, and others who's screens I have calibrated, much prefer the calibration when not targeting a white point or (shudder) a brightness. Targeting a brightness with the calibration for viewing video is simply a bad idea. There is no way you can notice a small change in brightness but you reduce saturation on all colors in order to target a brightness, which reduces contrast too. The only way to reduce brightness with a calibration is to have the video card output 254 or less when the source pixel's value is 255, this requires lowering all values to avoid banding, reducing the dynamic range for no visible benefit.
[...]
This may give the impression that I think calibrations are somewhat useless but the calibrated gamma curve is generally much better, as long as you do not get banding, and tuning the white point with the meter and the monitor's controls often helps a lot.Good points, Asmodian. It seems to me any unnecessary altering of an 8-bit video signal should be avoided. I am surprised calibration is not always done at the monitor, and in the analog domain - ie DAC gain & offset. This is preferable I think - except, as you say, for print work and such: https://en.wikipedia.org/wiki/Color_calibration#Display
Depending on the calibration software and type of monitor used, the software either creates a correction matrix (i.e. an ICC profile) for color values before being sent to the display, or gives instructions for altering the display's brightness/contrast and RGB values through the OSD.


Screen shots don't show us anything, they look different on my screen than they do on your's. ;)Yep! Another thing, calibration at the GPU messes up the colors if you are capturing the screen.
https://www.google.com/search?q=fraps+yellow+tint