or Connect
AVS › AVS Forum › Display Devices › Display Calibration › A comparison of 3DLUT solutions for the eeColor box
New Posts  All Forums:Forum Nav:

A comparison of 3DLUT solutions for the eeColor box - Page 4

post #91 of 603
@spacediver

Many thanks my friend for explaining in such detail, I appreciate your patience and effort within the reply.

It would seem that I was somewhat optimistic in even considering the possibility that the geometrically regular grid was even approaching the variation the human eye can detect.

I now understand the need for supplementary targeted points to complement the existing regularly spaced ones, none of which are surplus to requirements.
post #92 of 603
One thing that gets lost in these sorts of discussions is proportionality and diminishing returns. Has anyone done any semi-objective tests to determine how much additional benefit is gained as the number of calibration points increase?

For example, it is uncontroversially true that a 5^3 matrix gives better results throughout the color space than a standard WRGBCYM color calibration. If a display requires LUT calibration (not all do), the difference between the two is measurably significant. However, how much added benefit is gained by going up to 9^3 and then to 17^3, and finally up to 21^3?

Common sense suggests that most if not all of the low-hanging fruit will have been picked with 5^3. I wonder if anyone can quantify the degree of benefit offered by each successive increase in color space sampling and correction?
post #93 of 603
Quote:
Originally Posted by TomHuffman View Post

Common sense suggests that most if not all of the low-hanging fruit will have been picked with 5^3. I wonder if anyone can quantify the degree of benefit offered by each successive increase in color space sampling and correction?

I imagine that this largely depends on how linear the display is.
post #94 of 603
Thread Starter 
Quote:
Originally Posted by TomHuffman View Post

One thing that gets lost in these sorts of discussions is proportionality and diminishing returns. Has anyone done any semi-objective tests to determine how much additional benefit is gained as the number of calibration points increase?

For example, it is uncontroversially true that a 5^3 matrix gives better results throughout the color space than a standard WRGBCYM color calibration. If a display requires LUT calibration (not all do), the difference between the two is measurably significant. However, how much added benefit is gained by going up to 9^3 and then to 17^3, and finally up to 21^3?

Common sense suggests that most if not all of the low-hanging fruit will have been picked with 5^3. I wonder if anyone can quantify the degree of benefit offered by each successive increase in color space sampling and correction?

I tried to put things in perspective in the OP, for example on my own display which has a maximum gamut error of dE2000 = 8 in native mode.

  1. To get 95% of gamut errors below dE=3 requires only the internal WRGBCMY controls.
  2. To get ~70% of gamut errors below dE=1 requires either a 17^3 cubic grid or the CalMAN procedure.
  3. To get 93% of gamut errors below dE=1 requires 2500 optimally placed measurements.

I could imagine that a few hundred would put the display somewhere between 1 and 2.
Edited by zoyd - 2/19/14 at 5:27pm
post #95 of 603
I think Silly Sally's review of his use of Zoyd's patch sequence compared to a 21^3 sequence really says a lot.
The measurement metrics really do not tell the full story, which is understandable.

We have just supplied our beta testers with alternative optimised patch sequences, one with 2331 total patches, but that is equivalent to a 21^3 profile, and a 7470 patch sequence that is equivalent to a cube over 29^3 in size.

There are alternative optimised sets coming, but al are based on the basic fundamental that all colour must be treated equally, within an equal understanding of what calibration aims to achieve.

What I mean is calibrating a display is not the same as saying we perceive these colours this way, and those like that...
For accurate calibration all colours really must be treated equally, as you never know what colour conurbation will be shown by any given image.

Steve
post #96 of 603
Quote:
Originally Posted by TomHuffman View Post

For example, it is uncontroversially true that a 5^3 matrix gives better results throughout the color space than a standard WRGBCYM color calibration. If a display requires LUT calibration (not all do), the difference between the two is measurably significant. However, how much added benefit is gained by going up to 9^3 and then to 17^3, and finally up to 21^3?

One bit that should not be lost in the conversation, is the difference between calibration/profile points and the number of correction points for the hardware.

17^3 is just really barely enough to do correction with out trilinear interpolation causing significant artifacts. The testing in this case was done on a 64^3 device and all of the applications use the extra control points to help alleviate the artifacts of the hardware filtering. Lumagens only get away with such small cubes because they don't do simple trilinear interpolation . Early in the development of their cubes they found that using just trilinear interpolation on such a sparse cube caused show stopping amounts of error.
post #97 of 603
Quote:
Originally Posted by Light Illusion View Post


What I mean is calibrating a display is not the same as saying we perceive these colours this way, and those like that...
For accurate calibration all colours really must be treated equally, as you never know what colour conurbation will be shown by any given image.

Steve

Steve, can you answer this simple question:

Suppose that in advance, you decide to limit yourself to two thousand points.

Are you suggesting that it is better to spread those points regularly throughout the RGB space, as opposed to taking into account perceptual non uniformities?
post #98 of 603
Quote:
Originally Posted by TomHuffman View Post

One thing that gets lost in these sorts of discussions is proportionality and diminishing returns. Has anyone done any semi-objective tests to determine how much additional benefit is gained as the number of calibration points increase?

For example, it is uncontroversially true that a 5^3 matrix gives better results throughout the color space than a standard WRGBCYM color calibration. If a display requires LUT calibration (not all do), the difference between the two is measurably significant. However, how much added benefit is gained by going up to 9^3 and then to 17^3, and finally up to 21^3?

Common sense suggests that most if not all of the low-hanging fruit will have been picked with 5^3. I wonder if anyone can quantify the degree of benefit offered by each successive increase in color space sampling and correction?

Yes Tom, I started doing 5^3 for my Lumagen Mini and saw the improvements over a normal 6 point CMS.
Then saw a better way to go doing larger profiles (17^3) by using the external eecolor , once more a nice improvement.
Then Lumagen came out with the 20xx line that would do a about 729 point LUT using a larger profile like a 17^3. I did this because of what sotti is saying about the Lumagen. However even with the 2041's processing it was a step backwards from eecolor and a 17^3 profile..
Now going from a 17^3 profile to a 21^3 profile there is a slight improvement.

The improvement is about the same as going from a I1Pro 2 to a Jetti 1211 or going from zoyd's 2501 patch set or the Lumagen 2041 729 point LUT to a 21^3 eecolor profile/LUT.

The higher number in the patch set if well done, the more natural/even/refined the colors will be from one Blu Ray disc to the next. And it is at this level of accuracy the Jetti 1211 will shine as a reference meter for a K10-A.

So to me the question is, at what point is a acceptable level of accuracy.. Also how will the 4K technology play into all of this.

ss
post #99 of 603
Quote:
Originally Posted by zoyd View Post

I tried to put things in perspective in the OP, for example on my own display which has a maximum gamut error of dE2000 = 8 in native mode.
Of course, the D8000 also has a Movie mode whose gamut is MUCH more accurate than an average CIEDE2000 of 8.0.

It is really interesting how this field has changed over the years. I got really interested in display calibration around the time the JVC RS1 was released. Everyone seemed to love it because of its high contrast. I hated the the thing because it had color errors in the 11-12 dE range and no CMS. It almost single-handedly provided a raison d'etre for the first consumer external CMS, the original Lumagen Radiance. So in less than 10 years we have gone from removing 10 dE of error to obsessing over, maybe, 1.0 dE of residual error.
post #100 of 603
Thread Starter 
Quote:
Originally Posted by TomHuffman View Post

Of course, the D8000 also has a Movie mode whose gamut is MUCH more accurate than an average CIEDE2000 of 8.0.

That was maximum, average for the initial condition was 4.3. Average/max for the device using movie mode with the internal CMS is 1.2/3.6 dE2000
post #101 of 603
Quote:
Originally Posted by zoyd View Post

That was maximum, average for the initial condition was 4.3. Average/max for the device using movie mode with the internal CMS is 1.2/3.6 dE2000


zoyd,

Maybe you just have the wrong TV. With the VT60 in Pro mode and using its internal CMS, it is a snap to get an avg/max of 0.55/0.9 dE2000. smile.gif

Larry
post #102 of 603
Quote:
Originally Posted by zoyd View Post

That was maximum, average for the initial condition was 4.3. Average/max for the device using movie mode with the internal CMS is 1.2/3.6 dE2000
Sorry for confusing maximum with average. However, the point is the same. 1.2 dE is a very low color error for which I would generally not be too interested in attempting further correction. It would be nice if you could post dE data on the Movie mode post internal CMS for your D8000.
post #103 of 603
Quote:
Originally Posted by LarryInRI View Post

zoyd,

Maybe you just have the wrong TV. With the VT60 in Pro mode and using its internal CMS, it is a snap to get an avg/max of 0.55/0.9 dE2000. smile.gif
Yea, I had about the same result with a VT60 last summer.

http://www.avsforum.com/t/1174468/the-official-chromapure-thread/4600_100#post_23654807

The point I made at the time was that some displays are so good after using the internal controls correctly that 3D LUT correction of any type simply isn't necessary. Obviously, other displays are worse and clearly benefit. The Sony 4K 1000 projector also has nearly flawless color in Reference mode and, unlike the Panasonic, it doesn't even have a CMS.
post #104 of 603
Quote:
Originally Posted by TomHuffman View Post

Yea, I had about the same result with a VT60 last summer.

http://www.avsforum.com/t/1174468/the-official-chromapure-thread/4600_100#post_23654807

The point I made at the time was that some displays are so good after using the internal controls correctly that 3D LUT correction of any type simply isn't necessary. Obviously, other displays are worse and clearly benefit. The Sony 4K 1000 projector also has nearly flawless color in Reference mode and, unlike the Panasonic, it doesn't even have a CMS.

Hi Tom,

Generally, are the more expensive displays likely to be in the 'no need for 3D LUT' category with cheapo ones (like my Sharp Edge Lit LED LCD) probably benefiting?
post #105 of 603
Quote:
Originally Posted by spacediver View Post

Steve, can you answer this simple question:

Suppose that in advance, you decide to limit yourself to two thousand points.

Are you suggesting that it is better to spread those points regularly throughout the RGB space, as opposed to taking into account perceptual non uniformities?
It would be better to use a 'relatively' regular point set, either with a weighting or an optimised distribution.

What you would not want to use is a pseudo random point set with gaps in the coverage.

This is the basis of the optimised sets we have just released to our beta testers - with more versions being developed.
(We have been working on these sets for some time...)

Steve
post #106 of 603
Quote:
Originally Posted by Light Illusion View Post



What you would not want to use is a pseudo random point set with gaps in the coverage.

As zoyd showed in a previous post in this thread with two diagrams, the gaps are only relative to a perceptually non uniform space.

What you are suggesting would create gaps in a perceptually uniform space, and this is something that is not good. You want to sample the perceptual space uniformly. And to do that, you have to take into account human perception.

This is similar to the idea of delta E, which takes into account human perception.
post #107 of 603
Quote:
Originally Posted by TomHuffman View Post

Yea, I had about the same result with a VT60 last summer.

http://www.avsforum.com/t/1174468/the-official-chromapure-thread/4600_100#post_23654807

The point I made at the time was that some displays are so good after using the internal controls correctly that 3D LUT correction of any type simply isn't necessary. Obviously, other displays are worse and clearly benefit. The Sony 4K 1000 projector also has nearly flawless color in Reference mode and, unlike the Panasonic, it doesn't even have a CMS.

Disagree. You may want to do some profiling/LUT's.
Take a look at what I came up with using the posted patch set in this thread. http://www.avsforum.com/t/1517849/a-comparison-of-3dlut-solutions-for-the-eecolor-box/30#post_24367983

Also understand that I didn't use Green or any Gamma gains adjustments, in the above LUT.

Yes it very easy to come up with great looking charts for a standard calibration on a VT60 as it is for a 9G Kuro,, but without seeing how you got there leaves room for doubt on your statement.

btw, did you use your 1201 on the VT60. smile.gif

ss
post #108 of 603
Quote:
Originally Posted by PE06MCG View Post

Hi Tom,

Generally, are the more expensive displays likely to be in the 'no need for 3D LUT' category with cheapo ones (like my Sharp Edge Lit LED LCD) probably benefiting?
I don't think it has anything to do with cost. It just depends on how well the display is engineered. The Panny plasma I mentioned costs a little over $2K. You'd be surprised how poor the correlation is between cost and performance among consumer displays.
post #109 of 603
Quote:
Originally Posted by sillysally View Post

Disagree. You may want to do some profiling/LUT's.
Take a look at what I came up with using the posted patch set in this thread. http://www.avsforum.com/t/1517849/a-comparison-of-3dlut-solutions-for-the-eecolor-box/30#post_24367983

Also understand that I didn't use Green or any Gamma gains adjustments, in the above LUT.

Yes it very easy to come up with great looking charts for a standard calibration on a VT60 as it is for a 9G Kuro,, but without seeing how you got there leaves room for doubt on your statement.
You disagree with what, exactly? Post-calibration I measured the ColorChecker colors and a pri/sec saturation sweep inside the color space and the dEs were all below 1.0. You can't correct something that is accurate to begin with. If the color performance of a display is engineered well enough then getting the pri/sec colors, grayscale, and gamma right will result in great performance throughout the gamut. Now obviously a lot of displays are not well-engineered so this is necessary, but sometimes it isn't.

A client just ran a 729 color correction on the 4K Sony 1000 and the average pre-calibration dEs--the dEs before anything was adjusted in the color at all--on the full 729 set was 1.0. After the calibration was finished the average dE was 0.5. That's a waste of time.
post #110 of 603
Quote:
Originally Posted by TomHuffman View Post

A client just ran a 729 color correction on the 4K Sony 1000 and the average pre-calibration dEs--the dEs before anything was adjusted in the color at all--on the full 729 set was 1.0. After the calibration was finished the average dE was 0.5. That's a waste of time.

Agreed. I currently don't bother with colour calibration on my 1000es for that very reason.
post #111 of 603
Thread Starter 
Quote:
Originally Posted by TomHuffman View Post

Sorry for confusing maximum with average. However, the point is the same. 1.2 dE is a very low color error for which I would generally not be too interested in attempting further correction. It would be nice if you could post dE data on the Movie mode post internal CMS for your D8000.

The data summary for both native and CMS calibrated movie mode (labeled internal CMS) are in the first post.

post #112 of 603
Quote:
Originally Posted by spacediver View Post

you have to take into account human perception.
Actually, no, when performing accurate calibration human perception should not be taken into account, as that changes with the colour mix being displayed at any given time.
(As well as environmental considerations)

And as you never know in advance what colour will be shown in what combination you can't use that human perception as a basis for calibration.

This is why accurate calibration is done to a set of target values that encompass total volumetric accuracy..

You really do have to treat all colours equally, or you will never be sure of the final calibration.
post #113 of 603
Thread Starter 
Quote:
Originally Posted by spacediver View Post

As zoyd showed in a previous post in this thread with two diagrams, the gaps are only relative to a perceptually non uniform space.

What you are suggesting would create gaps in a perceptually uniform space, and this is something that is not good. You want to sample the perceptual space uniformly. And to do that, you have to take into account human perception.

This is similar to the idea of delta E, which takes into account human perception.

As you may have noticed, Steve engages in what's called magical thinking. This gives him super-duper immunity from facts and the conclusions of logical arguments that they lead to.
post #114 of 603
I don't understand this, but I just took the original LightSpace and Argyll LUTs as supplied by Zoyd, and applied them to a test image.

These are the results.

LightSpace LUT.


And the Argyll LUT
post #115 of 603
Quote:
Originally Posted by zoyd View Post

As you may have noticed, Steve engages in what's called magical thinking. This gives him super-duper immunity from facts and the conclusions of logical arguments that they lead to.
Or, to put it another way, Steve has a lot of experience from calibration in the real world, and has calibrated many 100's of displays for many hight-end post-production facilities and studios around the globe, and has tried just about all possible approaches to calibration over the years, including all the various approaches being discussed here, and has worked-out from practical experience what works and what doesn't...

Rather simple, I'd say biggrin.gif
post #116 of 603
Moderator

to the OP: there is no need to disparage another member: let's move on please
post #117 of 603
Quote:
Originally Posted by Light Illusion View Post

You really do have to treat all colours equally, or you will never be sure of the final calibration.

To be able to evaluate your equality, you need to be able to quantify it.

What colorspace are you going to quantify your equality in Steve, and why ?

Device RGB ?
Y,x,y ?
XYZ ?
L*u*v* ?
L*a*b* ?
DIN99 ?
Something else ?

Note that treating equally in one space is treating un-equally in another space.

Choose a space that has a poor correlation to what is visible, and you'll waste lots of sample points in volumes where it will make no visual difference, and you'll under sample volumes elsewhere that are the most visually inaccurate.

But why do you even think the equality of the samples in any space is a desirable property of a sample set ?

Hint - it's not. The goal is actually to reconstruct as visually accurate a model of the device behavior as possible from a sample set of a given size.
If the device behavior is well matched by the shape of the model, then few samples are needed.
If the match is less good, then more samples will be needed.
The nature of the model will determine what range of device responses it models "easily" (ie. with few samples) and what it models less "easily" (ie. more samples needed).

So an efficient sample set depends not only on what is visually significant, but also on the nature of the model and the way it is constructed from the sample set.
post #118 of 603
Display devices are not perceptual, they treat all colours the same - and as it is the display we are calibrating, not the uses perception of the display, that is what is important.

I agree 'optimised' patch sets are a good thing, but they need to be uniformly optimised.

It's why we have so many different profiling capabilities within LightSpace, with more coming.

A 'perfect' display can be calibrated with a grey scale and single RGB patch profile for example - a very optimised patch set.
But such displays are very rare, especially in the consumer world.

If you first perform a full volumetric profile of a display you could use that data to then define a very optimises set of patches to perform a very small patch set calibration.
But as you would have to do a full volumetric profile first, why not use that data to perform the calibration directly?
It would be far quicker in the long run.

So, I'm not disagreeing, I'm just trying to explain the real world.

Steve
post #119 of 603
Right, I worked out the very strange effect in the Argyll LUT - it cant deal with values over 235, and as most signal sources do clip at 16, but allow over-range peaks above 235 that could be a problem - but easy to fix.

So, having removed that error, this is what we are left with.

Try saving the two images to your PC and comparing them in a flip-flop mode, switching directly between them.
The artefacts in the Argyll LUT can be seen easily - and no, they are no 'fixing' issues with the display.

LightSpace LUT


Argyll LUT


Hint - look in Magenta and Blue
(Yes there are compression artefacts too, but that's not the LUTs)
post #120 of 603
Quote:
Originally Posted by TomHuffman View Post

You disagree with what, exactly? Post-calibration I measured the ColorChecker colors and a pri/sec saturation sweep inside the color space and the dEs were all below 1.0. You can't correct something that is accurate to begin with. If the color performance of a display is engineered well enough then getting the pri/sec colors, grayscale, and gamma right will result in great performance throughout the gamut. Now obviously a lot of displays are not well-engineered so this is necessary, but sometimes it isn't.

A client just ran a 729 color correction on the 4K Sony 1000 and the average pre-calibration dEs--the dEs before anything was adjusted in the color at all--on the full 729 set was 1.0. After the calibration was finished the average dE was 0.5. That's a waste of time.

What I am talking about is when doing a standard 10 point GS and a 6 point CMS. How you go about getting to the point you have good dE's, gamma ect. Not what your charts look like, what are you doing with the internal settings is what I am talking about.
Just because you can get good looking dE's, gamma, gamut, doesn't always mean you are getting the best performance from that display.

imo the charts I posted for Graeme's patch set using LightSpace are very good looking and show great results. However I have better performing LUT's .that don't have as good of looking charts and dE's.

Why do folks post when using say CM auto cal, that the dE's and charts look good (well within the 3 dE limits) but there still is a problem.

What did the client say, was there a Lumagen 20XX involved in this LUT, from were did this patch set come from. However all that makes no difference because we are talking about a VT60 and what your claim is.

ss
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Display Calibration
AVS › AVS Forum › Display Devices › Display Calibration › A comparison of 3DLUT solutions for the eeColor box