| Forums | Register | Polls | Search | Statistics |
You must be logged in to post content on this forum.
Display Calibration Light Illusion Forums / Display Calibration /  

Demoing LUT generation

Author NaN
#1 | Posted: 2 Dec 2022 01:58 
I just bought ColourSpace DPS for manual calibration of my monitors and projector.
Manual calibration works as expected - great!
As DPS also comes with a "demo" (no export) of the 3DLUT generation capability I wanted to check how much this improves the result compared to pure calibration. But somehow the results get worse, not better. I probably do things the wrong way...
Background: For this experiment I tried to calibrate my wide gamut QLED monitor for sRGB profile.
Applied steps:
* manual calibration with DPS and the global rgb sliders for white point and then the per-color hue/saturation sliders for primary and secondary colors until the results don't get better anymore that way. Much better closer to sRGB than factory settings but but still off by 8/10/1 dE for red/green/blue.
* run display characterization in "Grey, Primary & Secondary Ramp+" mode (so there are enough samples for all LUT tool modes later)
* save the "New" profile (well, save in RAM, DPS doesn't save to disk)
* open the LUT tools, select sRGB as Source profile and "New" as destination profile.
* generate one LUT each with the different modes (Hybrid, Map Space, Fit Space, Peak Chroma and Peak Luma)
* select one of the generated LUTs as Active LUT in the Profiling Settings and rerun display characterization

Now the the primary colors are visibly further away in the CIE diagram compared to the just-calibrated (no LUT) variant, and dE values are also much worse (e.g. 27/20/20 for red/green/blue). I also tried to reread min/max/contrast values after selecting the LUT - didn't make things better.

Afterwards I tried to greate profiles with DisplayCAL/ccProfile (for the manually pre-calibrated display but without LUTs - the same starting point as for the LUT experiments). There results got a little better - not much (still ~ 5dE off) but at least into the right direction. (This is probably a difficult case as this display was apparently not made with sRGB in mind but for at least DCI-P3...)

What am I doing wrong with the LUT generation?

Author Steve

#2 | Posted: 2 Dec 2022 10:25 
Unfortunately, it is impossible to even guess what you have done wrong with your LUT workflow.
You will need to supply screen-grabs of each step, and more information, etc.

But, a LUT can never increase gamut/saturation.
Now the the primary colors are visibly further away in the CIE diagram compared to the just-calibrated (no LUT) variant

That is just impossible.

Steve Shaw
Mob Boss at Light Illusion

Author NaN
#3 | Posted: 2 Dec 2022 12:09 
Sorry instead of "further away" I should have written "more off" (=higher deltaE), not wider gamut.
Will try to provide screenshots later on.

Author NaN
#4 | Posted: 2 Dec 2022 18:29 
Ok, have redone the experiment with snapshot documentation
This time the deltaE values didn't go through the roof as much as last time. The only difference: This time I used deltaE type 2000 instead of 1976.
But still, the LUT doesn't seem to improve the situation, it gets slightly worse in most cases and stays about equal in the best case.

But step by step:
First I reset my wide color gamut QLED monitor (LG 38WN95C-W) to factory settings and first just dialled down the brightness until I have about 100 nits (HDR mode in Windows is disabled, the factory ICC profile is selected).
Then I started ColourSpace on a separate notebook and the Calibration Client on the PC that is connected to the monitor via Displayport.
Here the settings and initial Characterization results:
1.1 Initial Settings
1.2 Probe Settings
1.3 Hardware Settings
1.4 Graph Options
1.5 Initial uncalibrated characterization
1.6 Initial White measurements
1.6 Initial Red measurements
1.6 Initial Green measurements
1.6 Initial Blue measurements
dE Values for secondary colors (if needed I can provide the screenshots here also):
Magenta: 3.3935
Cyan: 4.6923
Yellow: 3.7010

Afterwards I manually calibrated the display by using the global Rgb sliders and per color hue and saturation sliders. These were the results:
2.1 White after manual calibration
2.1 Red after manual calibration
2.1 Green after manual calibration
2.1 Blue after manual calibration
Magenta: 2.4369
Cyan: 3.6730
Yellow: 0.5534

This profile was saved as "AfterManualCalibration"

Now I created two LUTs with mappings from sRBG to the AfterManualCalibration profile, with the MapSpace and PeakChroma algorithms:
3.1 Create LUT with MapSpace algorithm
3.2 Create LUT with PeakChroma algorithm

I don't want to waste our time with trying all permutations, but I tried PeakChroma with the original min/max contrast settings and MapSpace with remeasuring min/max contrast - not sure what the expected workflow would be. (Probably the latter makes more sense?).

So first I set the Active LUT to the PeakChroma LUT now:
4 Set active LUT to PeakChroma one
Here the resulting colors (after another characterization run):
4 White for PeakChromaLUT
(It appears plausible to me that the dE for white is now worse because the min/max contrast cannot be reached anymore with the LUT?)
4. Red for PeakChromaLUT
4. Green for PeakChromaLUT
4. Blue for PeakChromaLUT
Magenta: 4.4466
Cyan: 6.2120
Yellow: 6.1341

And finally the same game with the MapSpace LUT - this time with the min/max contrast remeasured:
5. MapSpace LUT settings
5. White with MapSpace LUT settings
5. Red with MapSpace LUT settings
5. Green with MapSpace LUT settings
5. Blue with MapSpace LUT settings
Magenta: 0.7805
Cyan: 6.7194
Yellow: 5.9318

My approach was proably naive because I manually calibrated the display to the best possible dE values I could - perhaps it might have been better to leave some saturation margin for the LUT algorithm to work? But I wanted to have the results without LUT as good as possible, because many programs won't use ICC profiles, and those should also get good results. (And currently I cannot save and export to ICC with my ColourSpace DPS version anyways...)
But I would at least have expected that the LUT algorithm when it realizes that it cannot make things better at least doesn't make things worse?
(Unfortunately I don't get the results from yesterday anymore - there the dE values got worse up to dE old value+10 - and not better in any case).
Or did results actually get better and dE is just the wrong metric to measure that? Or perhaps my settings were just wrong...

Thanks for reading down to that point - thanks for having a look!

Author NaN
#5 | Posted: 2 Dec 2022 20:01 
Did another experiment: I reset the monitor to factory settings again and this time I only manually calibrated brightness and white point and left all the saturation and hue values at default settings.

That is what I got after this initial minimal manual calibration:
White: 1.5083
Red: 3.8914
Green: 4.9392
Blue: 0.8113

When generating a LUT (PeakChroma) based on these setting, I get much better results. Here with leaving the original min/max contrast from above in place:
White: 0.3968
Red: 0.2883
Green: 0.3569
Blue: 0.5966

And here remeasuring min/max contrast after the LUT was set to active:
White: 0.1698
Red: 0.1317
Green: 0.0958
Blue: 0.7242

So the approach of choice seems to be to do only minimal manual calibration and to take the post-LUT min/max contrast values.
This is good! Unfortunately it would mean that one cannot calibrate to as-good-as-it-gets manual values - so everything that ignores ICC profiles still looks good - and then having an ICC profile that makes things even slightly better for the programs that interpret it. That said, the non-ICC variant isn't that bad for non color critical things (and color critical stuff should have ICC support...)

And it wouldn't work with my current ColourSpace license. Am I right that the least expensive variant to create ICC profiles with this software would be ColourSpace HTL+SpaceMan ICC? That's quite a step up. Perhaps it would be an idea to consider adding just ICC export to the DPS version? Then all the LUT stuff wouldn't be a mere demo and the pro-stuff that relies on specific 3dLUTs would still be reserved for the bigger variants?

Well in a first step I think I'm glad that I (hopefully...) have understood the (surface of the) approach now and can stop wondering
Perhaps the LUT algorithm could be made aware of cases where some preconditions are not met and thus the results would be worse than the starting point? Getting at least a warning and a hint what to change might be helpful!

P.S.: Tried to generate a similar ICC with ccProfile and DisplayCal and got worse results than with ColourSpace: Mainly the white point got further corrected the color gamut almost not. Thing is - I measured with ColourSpace (remotely over the Calibration Client). Could it be that the Calibration Client is not ICC aware and thus the gamut part of the ICC profile that is loaded in the OS is ignored when doing Characterization with ColourSpace? (Which could explain why the white point is applied?)

Author Steve

#6 | Posted: 2 Dec 2022 21:32 
If there is an active ICC for the HDMI output to the display from the laptop, you will never get a good calibration.
You must disable all ICC profiles, as per the User Guide for direct HDMI profiling.

Steve Shaw
Mob Boss at Light Illusion

Author NaN
#7 | Posted: 2 Dec 2022 21:47 
When doing the LUT experiments the ICC profile was disabled (well, the factory-one chosen). The experiments with the ICC profiles from ccProfile&Co were done afterwards, as an attempt to get a ColourSpace Characterization of those profilings and to have a line of comparison with the ColourSpace generated LUT from the step before.
So having ICC support inside the Calibration Client would be helpful to characterize a display with ColourSpace with an existing ICC profile, compare ICC profiles etc. Or in order to see whether the display is well calibrated with the profile active or whether it needs a new ICC profile.
(Edit: Assuming we are talking about a PC monitor (like in this case) that is supposed to be used with a particular PC, so PC, ICC Profile and monitor would have to be seen as a unity when it comes to characterization whether this combo is color correct)

Author Steve

#8 | Posted: 2 Dec 2022 21:56 
You cannot have ANY active ICC profile - nothing.
The HDMI output MUST be clean, and bit-perfect accurate.
Please read the info in the link provided.

Steve Shaw
Mob Boss at Light Illusion

Author NaN
#9 | Posted: 2 Dec 2022 22:31 
Ok, thanks, good point! I understand that if I would export the LUT above and make a ICC profile from that and load it instead of the default one, the results wouldn't be what I would expect. Could one (instead of removing all ICC profiles which is kind of a hazzle in Windows as the only option for "no profile" seems to be to uninstall them all(?)) also merge (chain) the default ICC profile with the new one and thus could leave the default ICC profile (that gets automatically installed when plugging in a recognized monitor) active during calibration and profiling?

And I guess this doesn't necessarily apply for manual calibration?
Assuming I would manually calibrate the monitor with a clean input signal generator - and then reattach my usual PC (without fighting with it to not load the default ICC profile for that monitor ever again), colors would be off, right?
So, in the case of manual calibration, couldn't the PC with the standard ICC profile for that monitor be seen as yet another specific input signal source that should best be used during calibration in exactly the configuration that it will later on be delivering the images? I guess this would be the way of the least resistance

For the same reasons it is probably easier to have two separate profiles for the projector - one that gets callibrated with a disc of test images for the Bluray player and a separate profile for the notebook (then with whatever ICC profile that gets automatically installed once I plug in the projector for the first time)? Will have to check whether separate profiles for different inputs are possible with the projector (it's not there yet) - fingers crossed!

Author NaN
#10 | Posted: 5 Dec 2022 22:23 
Ok, my initial suspicion was true - the Calibration Client doesn't properly consider the active ICC profile. Thus it is not possible to use it to characterize the currently active profile.
In order to judge the color correctness of the ccProfile and DisplayCAL ICC profiles I created a document in Affinity Publisher (which does consider ICC profiles correctly) with seven pages - full screen white, one page each for the three primary and for the three secondary colors.
Then I measured that with Colourspace manual measure.

The results:
white: 1.1040
red: 0.2660
green: 0.4073
blue: 0.5295
yellow: 0.3786
cyan: 0.8440
magenta: 0.4690

white: 1.0401
red: 0.2534
green: 0.7734
blue: 0.5650
yellow: 0.5800
cyan: 1.0880
magenta: 0.4435

ColourSpace's LUT feature was a lot better in hitting the white point (but there I am not 100% sure whether DisplayCal and ccProfile perhaps have a different interpretation which white point to target for sRGB...). For the colors it is give or take.

My conclusion for the moment: For mere monitor calibration (of consumer level monitors that do not accept to upload native LUTs) the usual profiling software is good enough for me. The manual measurement features of ColourSpace DPS are awesome though. So I think I got the right edition for me

Thinking about adding ICC profile support to the Calibration Client would be nice to have - measuring color correctness without considering the profile is a little meaningless. And for LUT-based profiling it might be nice to have a feature to programmatically disable the ICC profile during measurement (instead of having to uninstall it) - like the other monitor profiling tools all offer.
(If this would be a feature of the main application and not the Calibration Client then it would be good if we could install the application on more than one PC - think copy protection is getting unfavourably in the way of usability in the case of ColourSpace... The imho best variant I have seen in other software would be a personalized keyfile that can be used without the need for an online activation at any number of computers. This approach allows blacklisting of leaked keys and hundreds of check-criteria that can be added one after the other in subsequent updates - so a keygen for version X would not work for version X+1 - each version would have to be cracked individually, same hacker-discouragement as for activation/dongles. Imho the best balance between user friendlyness and copy protection!)

Author Steve

#11 | Posted: 5 Dec 2022 22:32 
For ICC based calibration, the use of SpaceMan is required.
It takes 3D LUTs and converts them into ICC profiles.

We would never recommend profiling through an existing ICC profile, as the website states.
There are far too many inconstancies and issue with them.
And that is why TPGs are not 'ICC Aware'.
Not the inbuilt one in ColourSpace, nor the separate Calibration Client, nor any others...
(And that ignores the fact that ICCs are not 'OS wide' in Windows.)

For some understanding of ICC issues, see: What's wrong with ICCs?

Steve Shaw
Mob Boss at Light Illusion

Author NaN
#12 | Posted: 5 Dec 2022 23:02 
Thanks for the link, interesting read! Monitors that accept native 3D LUTs are probably out of my price range - and mere calibration with the inbuilt CMS alone (at least for my monitor) is not very precise - the addition of a fitting ICC profile can still improve the calibration results for me as shown above. So I guess I'll stay with the manual pre-calibration (of at least white point+brightness+contrast)+ICC profile route for my existing monitors.

But I guess that ColourSpace is aiming much higher - professionals who need the highest color accuracy they can get, will probably buy such a 3DLUT capable display - and then the higher versions of ColourSpace can be used at full capacity!
Have only scratched the surface of what is possible with these LUT features - but looks powerful and intuitive so far! But I think I know what I wanted to learn from this experiment now - my current hardware doesn't justify the "bigger" LUT-generation capable versions. Have used ColourSpace DPS now to manually calibrate (and with the software that comes with the Colorimeter create ICC-profiles for) all my monitors and am looking forward to manually calibrate my new projector with DPS - for which it should be optimal!

You must be logged in to post content on this forum.
Display Calibration Light Illusion Forums / Display Calibration /
 Demoing LUT generation


Online now: Guests - 1
Members - 0
Max. ever online: 192 [11 Jan 2023 08:39]
Guests - 192 / Members - 0