2015-10-19 at 19:19 #1683
Now I don’t know about LCD displays, may be a whole different story, but for CRT and plasma displays, why can’t we measure, say, 254 levels of R, G and B separately and be done with it (aka have the perfect profile)? After all, normally we won’t get better than that. Or can we? The only reason I can think of is that phosphors somehow influence each other, i.e. if G and B are lit together then the resulting formula won’t be simly G+B luminance but something else. What’s really going on in here?2015-10-19 at 19:30 #1684
This depends on the display. You can get a really good profile from just a few dozen patches if the display is perfectly additive and the internal display electronics (processing etc) don’t screw it up. Lowest patch amount you can set in dispcalGUI via the slider is 73. A simple shaper+matrix profile created from those 73 readings gives me an average dE of 0.5 and max of 2.2 (~700 patches used for verification), which is really good, on a NEC LCD2690WUXi (older LCD CCFL). I usually do a 596-patch LUT profile though because that gives me an average of 0.3 and max of 1.6.2015-10-19 at 20:17 #1685
So basically we need to account for imperfect additivity? Is that even possible with non-LCD displays? Electronic issues aside…
Sidetracking the discussion, I’m trying to create the perfect profile for the Kuro, but it seems to create slightly different profiles every time I make a run. Curves for example may look more saturated than LUT and it’s really hard to understand which one is right. It just cannot be repeated. The strange thing is that running a verify pass wouldn’t show any major differences between the two profiles.2015-10-19 at 20:18 #16872015-10-19 at 20:22 #1689
The problem is, I cannot make it repeatable. Florian, Can you share some advice on how to create the perfect profile by hand, manually? I think it’s the display itself acting up as I’ve tried two different measurement devices and they behave the same. Automatic settings just don’t work with this panel…
Even though I have physically adjusted the color temperature to be 5700 now, the profile that was created with 6500 in mind looks better. Here are the results of two test runs on two profiles above.2015-10-19 at 20:22 #16912015-10-19 at 20:37 #1693
Sorry, I actually only changed the monitor profile and didn’t change the profile in dispcalgui. Running it with a different dispcalgui profile yields a very different result as expected. Still, this doesn’t help with figuring out the right workflow for calibrating tricky displays.2015-10-20 at 9:15 #1694
Any reason you’re using the shaw & fairchild 1997 observer instead of the default?
Things I would do:
2015-10-21 at 19:08 #1695
- Disable black level drift compensation (not necessary for colorimeters generally)
- Disable 1D LUT calibration (set tone curve on the “Calibration” tab to “As measured”)
- Reduce the number of test points (aim for around ~1500-3000 for LUT profiles, 73 for shaper + matrix)
- Make sure the test window is small enough to not trigger ABL
I already tried most of the bullet points that you mentioned, drifts were completely disabled, tried different number of test points and of course set the window to the smallest size at the same time making the rest of the display black and making sure nothing (popping up windows and ambient light) could interfere with the measurement process.
One thing I tried just a couple of times was setting the tone curve to “as measured”.
So, I’ve performed an additional batch of profiling runs, curves/matrix with various number and combinations of patches and LUTs with different numbers of patches. The net result is that LUT profiles regardless of the number of patches give me better “verifiable” results, whenever I try an ISO12646 test, curves+matrix perform much worse numbers-wise. But! All LUT profiles come out undersaturated and flat-looking compared to curve-based profiles. Now I know that going for saturation is not the right thing of course and that applying a generic wide gamut profile to a display will result in a totally undersaturated image, whilst applying a generic sRGB will do the reverse, et cetera, et cetera. In this particular case though I have a couple of printouts made by a professional laboratory in Germany and the original digital photos to compare. The printouts that they made are actually closer in saturation to curve-based profiles and there’s a very noticeable difference between them and those LUT profiles I generate. They didn’t apply any corrections to the images so I use them as a sort of reference.
Now the insane part:
1. the panel’s readouts seems to fluctuate depending on some random factor. I never try to calibrate right after turning it on, rather allow it to “warm up” for quite a bit of time doing some work. But one day I measure the white balance and it’s one thing. Next day/hour I try the same and the results seriously differ from the last time I brought all bars into the marked position. I adjust them again and the next time I have to adjust them again. It always goes back and forth.
2. The panel sometimes (but not always!) wants to gradually lower the brightness if a patch of white is displayed continuously. That sometimes happens during the interactive adjustment. That doesn’t affect the white balance.
3. If I try to display another patch of white/gray (or a generic window) beside the measurement window in interactive mode or make the measurement window even slightly bigger, white balance immediately changes: blue and red are reduced and green seems to stay the same. Which begs the question of whether the luminance of a patch can affect the final result. For example, I start with a window and a black background and get: x: 0.3118 y: 0.3282. I try a bigger window and get x: 0.3120 y: 0.3300 so I go from deltaE 0.5 to deltaE 1.2. Changing the background to some color instead of gray would result in ever bigger deviations. Is there any way to tame this wild horse?2015-10-21 at 19:42 #1696
I also often get a table of dubious measurements at the end with varying amount of suspicious patches (from 2 to 16).2015-10-21 at 19:55 #1697
Seems like the display is not very stable, although that’s not out of the ordinary for a Plasma.2015-10-21 at 20:44 #1698
Yeah, but I was under impression one could still tame it somehow. I do get decent results working with photos on it, after all the printouts matched the display closely, especially allowing me to see fine color details. It’s just when you try to calibrate it it’s always somewhat off in one way or another. Maybe I could generate some specific set of LUT patches that would help with the undersaturation issue? I looked into it but don’t really know if any specific parameter is going to influence this particular situation.
Also, what does it mean when the curves-based profile is so different from LUT-based one? Unstable additivity?2015-10-21 at 21:14 #1699
Also do you think that if I routinely pass verification with a certain profile that it more certainly than not indicates that it is actually valid? Just now I did a run of CMP Digital Target 4 and here’s the report…2015-10-21 at 21:21 #1702
I looked into it but don’t really know if any specific parameter is going to influence this particular situation.
I doubt it as well. This seems to be a display-specific issue. I woould seek the help of someone who has already successfully used 3D LUTs with a Kuro. AVSforum could be a place for this.
Also, what does it mean when the curves-based profile is so different from LUT-based one? Unstable additivity?
All I can see from the profiles you attached is that the shaper+matrix one is not a good match to the measurement data (avg. dE76 of 3.29, max of 16.34), while the LUT profile is better although by no means as good as I would expect (avg 1.79, max 12.18 – I’m used to errors that are less than one third of that).2015-10-21 at 21:24 #1701
That certainly looks good.