A comparison of 3DLUT solutions for the eeColor box - Page 24 - AVS | Home Theater Discussions And Reviews
Baselworld is only a few weeks away. Getting the latest news is easy, Click Here for info on how to join the Watchuseek.com newsletter list. Follow our team for updates featuring event coverage, new product unveilings, watch industry news & more!


Forum Jump: 
 45Likes
Reply
Thread Tools
post #691 of 709 Unread 04-27-2015, 01:25 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
I recommended absolute colorimetric with white point scaling, but that's got nothing to do with my point. That's recommended whenever you can't pre-calibrate to D65.

Last edited by zoyd; 04-27-2015 at 01:51 PM.
zoyd is online now  
Sponsored Links
Advertisement
 
post #692 of 709 Unread 04-27-2015, 01:50 PM
AVS Special Member
 
sillysally's Avatar
 
Join Date: Jun 2006
Location: Chicago
Posts: 3,975
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 253 Post(s)
Liked: 340
Yes I also use absolute colorimetric with white point scaling, but I wasn't talking about that.

I am talking about White level drift compensation.

ss
sillysally is offline  
post #693 of 709 Unread 04-27-2015, 01:58 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
sorry ss, the comment was meant for reply to:

Quote:
Originally Posted by Light Illusion View Post
If you mean rendering intent, that was set to 'Absolute colorimetric', with Tone Curve Rec.1886.

I used the suggested setting you gave me, so all should be ok
If Steve didn't use white point scaling then he may not get the best results when starting from a non-D65 white point.
zoyd is online now  
post #694 of 709 Unread 04-27-2015, 02:03 PM
Advanced Member
 
Light Illusion's Avatar
 
Join Date: Aug 2010
Posts: 584
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 30 Post(s)
Liked: 255
Yes, I used absolute colorimetric with white point scaling - I was was just being lazy in my post, and not typing the whole line of text...

Steve

Steve Shaw
LIGHT ILLUSION

Light Illusion is online now  
post #695 of 709 Unread 04-27-2015, 02:12 PM
AVS Special Member
 
Citation4444's Avatar
 
Join Date: Aug 2004
Location: Georgia Mountains
Posts: 1,032
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 118 Post(s)
Liked: 64
Where do you set white point scaling?
Citation4444 is offline  
post #696 of 709 Unread 04-27-2015, 02:34 PM
AVS Addicted Member
 
Join Date: Sep 2002
Location: Wetumpka, AL
Posts: 15,882
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 452 Post(s)
Liked: 281
Quote:
Originally Posted by sillysally View Post
I did mention to you a few months ago about the White drift compensation for Plasma's. Glad you are going to perfect that option and add to LS.

ss
I did a 10,000 point profile on my F8500 plasma using Calman one night last week. At the end of the session, the pattern area was clearly offcolor. This white drift compensation sounds like something that should be in all LUTs programs.
sillysally likes this.

Tools: Calman 5, Chromapure, VideoForge HDMI II, Klein K10A, eecolor colorbox.
Displays: Samsung 64F8500, Panasonic 65VT50, Oppo 95, Pioneer 51, Tivo Roamio and Mini, Dish VIP722.
Audio: Marantz AV8801, Rotel Amps, Atlantic Tech 8200 speakers, Seaton Submersive HP
JimP is online now  
post #697 of 709 Unread 04-27-2015, 03:00 PM
AVS Special Member
 
sillysally's Avatar
 
Join Date: Jun 2006
Location: Chicago
Posts: 3,975
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 253 Post(s)
Liked: 340
Quote:
Originally Posted by Citation4444 View Post
Where do you set white point scaling?
Those settings are if you are using dispcalGUI.
White level drift setting can be found on Display and Instrument page.
Absolute colorimetric with white point scaling is on the 3DLUT page.

Quote:
Originally Posted by JimP View Post
I did a 10,000 point profile on my F8500 plasma using Calman one night last week. At the end of the session, the pattern area was clearly offcolor. This white drift compensation sounds like something that should be in all LUTs programs.
Yes its a nice feature when used to profile. But don't expect a big difference.

Yes I have seen the offcolor patch window when using a external pattern generator.
However when using madVR/TPG internal patch generator and White level drift, I have very little IR (offcolor) at the end of a very large profile.

ss

Last edited by sillysally; 04-27-2015 at 03:05 PM.
sillysally is offline  
post #698 of 709 Unread 04-28-2015, 06:13 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
Here is another test comparing regular grid to optimized patch performance using both LS and Argyll algorithms. This time I only used LS for both profiling and verification, so it's a direct algorithm comparison of each patch configuration. To do that I exported the LS measured profile and ran it through the Argyll tools to calculate the LUT.

Top plot shows the 14^3 grid based eeColor LUT verification results and the bottom one shows the results after profiling with the same number of patches optimized using ArgyllCMS. The LS results are better than Argyll using the grid set and vice-versa using the optimized set. In both programs the optimized set reduces color errors compared to the grid set with Argyll showing a larger difference between the two configurations.

ConnecTEDDD likes this.
zoyd is online now  
post #699 of 709 Unread Yesterday, 01:14 AM
Advanced Member
 
Light Illusion's Avatar
 
Join Date: Aug 2010
Posts: 584
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 30 Post(s)
Liked: 255
I suspect those differences, which really are just 'down in the noise', and not really worth spending time on, are just rounding differences with the maths used for the final LUT processing.

We very deliberately use processing that is very, very fast compared to all other systems, as you know

We would rather use the time to profile additional points, as in my first comparison, as that will always give superior results.
And with our speed benefit we can easily use double the profiling points, if not more

But thanks for the direct comparison, as that really does fit our thoughts.

What about the same comparison with other CMS systems?
That would be interesting too.

Steve
ConnecTEDDD likes this.

Steve Shaw
LIGHT ILLUSION

Light Illusion is online now  
post #700 of 709 Unread Yesterday, 03:52 AM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
Quote:
Originally Posted by Light Illusion View Post
I suspect those differences, which really are just 'down in the noise', and not really worth spending time on, are just rounding differences with the maths used for the final LUT processing.
I don't think they are from rounding but send me your source code and I'll have a look. The repeatability (noise) level of the histogram means is 0.02 dE00.

Quote:
We very deliberately use processing that is very, very fast compared to all other systems, as you know

We would rather use the time to profile additional points, as in my first comparison, as that will always give superior results.
And with our speed benefit we can easily use double the profiling points, if not more
The computational speed of LS is impressive. Argyll takes about 15 minutes to fit the profile and then do the gamut mapping on my midrange laptop. For retargeting purposes once the profile fit is saved, it takes about 7 minutes. LS takes about 15 seconds. Other than that difference, the process time is the same when using the same number of patches and the same probe at the same integration time settings between LS, Argyll, and CM.

Quote:
What about the same comparison with other CMS systems?
That would be interesting too.

Steve
I already have some data on that, current internal CMS calibrations can be pretty good if you spend a lot of time tinkering with them.


Last edited by zoyd; Yesterday at 06:07 AM.
zoyd is online now  
post #701 of 709 Unread Yesterday, 05:12 AM
Advanced Member
 
Light Illusion's Avatar
 
Join Date: Aug 2010
Posts: 584
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 30 Post(s)
Liked: 255
Yeah, 0.02 dE00 errors are seriously not worth the time and effort to chase - especially as the cost of time.
The one thing we hear from professional calibrators all the time is 'time is money'.

Or, many home calibrators would rather use the time to increase the patch set size.
Either is fine with us, and it the approach we take, and will,be sticking with.

On my small Laptop Argyll takes over 25 minutes to calculate the LUT - LS takes less than 10 seconds.
Profiling time is approximately the same, but again we would rather not use Low Light Averaging, but measure more patches.
The result of increased patch numbers is nearly always superior to Low Light Average measurements, as you have more data to work with, if you colour engine know how to deal with some 'bad' samples.

But, we do have to agree - we really are talking inconsequential error levels now

Steve Shaw
LIGHT ILLUSION

Light Illusion is online now  
post #702 of 709 Unread Yesterday, 05:40 AM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
Quote:
Originally Posted by Light Illusion View Post
Yeah, 0.02 dE00 errors are seriously not worth the time and effort to chase - especially as the cost of time.
The one thing we hear from professional calibrators all the time is 'time is money'.
You misunderstood my point, the error plots are repeatable to the 0.02 dE00 level, so that differences above that level between the mean error values in the graphs is real and not noise. I do agree that the differences shown above are only interesting from a mathematical point of view, and are not of any visual consequence. And it is true that you can always choose to run more patches to reduce those differences as you run up against the precision limits of the set-up you are calibrating. In my particular case running more then 2500 patches is a waste of time.

Last edited by zoyd; Yesterday at 06:08 AM.
zoyd is online now  
post #703 of 709 Unread Yesterday, 05:44 AM
Advanced Member
 
Light Illusion's Avatar
 
Join Date: Aug 2010
Posts: 584
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 30 Post(s)
Liked: 255
Sorry, I was using the term 'noise' in the way we Brits tent to use it - to mean 'of no consequence', not that is was actually 'noise'!

A good example of two countries being separated by a common language

Steve Shaw
LIGHT ILLUSION

Light Illusion is online now  
post #704 of 709 Unread Yesterday, 01:26 PM
AVS Special Member
 
sillysally's Avatar
 
Join Date: Jun 2006
Location: Chicago
Posts: 3,975
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 253 Post(s)
Liked: 340
Quote:
Originally Posted by zoyd View Post
I don't think they are from rounding but send me your source code and I'll have a look. The repeatability (noise) level of the histogram means is 0.02 dE00.

The computational speed of LS is impressive. Argyll takes about 15 minutes to fit the profile and then do the gamut mapping on my midrange laptop. For retargeting purposes once the profile fit is saved, it takes about 7 minutes. LS takes about 15 seconds. Other than that difference, the process time is the same when using the same number of patches and the same probe at the same integration time settings between LS, Argyll, and CM.

<<<Snip>>>
It takes me about 90 seconds to make 3DLUT for my eecolor using argyll. I will check on how long it takes to profile fit, but I am sure its much faster than 15 min.

Probably about 5 seconds using LightSpace.

ss
sillysally is offline  
post #705 of 709 Unread Yesterday, 01:47 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
wow, I guess I need to upgrade my laptop.


edit:

duh yeah, it helps when I'm not using 3 year old hardware. On my macbook pro the profile fit takes 60 seconds and the LUT mapping takes 90 seconds.

Last edited by zoyd; Yesterday at 04:09 PM.
zoyd is online now  
post #706 of 709 Unread Yesterday, 07:24 PM
Advanced Member
 
gwgill's Avatar
 
Join Date: Jan 2013
Location: Melbourne, Australia
Posts: 723
Mentioned: 7 Post(s)
Tagged: 0 Thread(s)
Quoted: 122 Post(s)
Liked: 104
Quote:
Originally Posted by zoyd View Post
Argyll takes about 15 minutes to fit the profile and then do the gamut mapping on my midrange laptop. For retargeting purposes once the profile fit is saved, it takes about 7 minutes.
It's actually the 1D input/output curves that take a lot of the time - the core fitting is quite fast on modern hardware compared to when I first wrote it. The inverted lookup during linking is quite slow - but not too bad for RGB compared to CMYK. The ArgyllCMS toolset isn't optimized for just a Video workflow, but is much more general purpose and covers a much greater range of devices and tasks. I've typically gone for accuracy over speed when choosing an approach, and I haven't attempted to parallelize any of the core code yet.

As an illustration - a "cheat" that can be used for RGB is to create a direct, inverted interpolation model of the device behaviour from the measurement points - i.e. CIE in and RGB out, or even Rec709 RGB in to Device RGB out. Doing that using ArgllCMS's core fitting engine with fixed or no per-channel curves would be very quick - a matter of seconds (very similar to the "colprof -ni -no" timing). You don't get general gamut mapping, you probably have to use some heuristics for clipping (which should be fine), and you won't detect any bad behaviour (non-monotonicity) in the display - the interpolation model will fudge over it, but that will probably be fine in most situations because the Video case is typically making relatively small corrections between two very similar spaces. But is the speed of profiling & link/3DLut creation very important in the end ? Most of the time will be spent waiting for sufficient measurements to be made, and you would hope you don't have to do it very often.
gwgill is offline  
post #707 of 709 Unread Yesterday, 10:14 PM
Senior Member
 
fhoech's Avatar
 
Join Date: Apr 2013
Location: Stuttgart, Germany
Posts: 496
Mentioned: 18 Post(s)
Tagged: 0 Thread(s)
Quoted: 208 Post(s)
Liked: 128
Argyll 1.7b6 processing times on a fairly old (but slightly overclocked, 3.1 -> 3.6 GHz) dual core desktop:

~2700 patches
Profile computation time: ~3 minutes
3D LUT (65^3 inverse linking): ~5 minutes
3D LUT (17^3 inverse linking): ~10 seconds

~10000 patches
Profile computation time: ~18 minutes
3D LUT (65^3 inverse linking): ~5 minutes
3D LUT (17^3 inverse linking): ~10 seconds

Computing B2A tables with low resolution (as in these tests) greatly benefits profiling speed.

When generating the 3D LUT, the more of the source gamut is within the destination, the speedier the process. In this case, the destination profile was from a display that covers roughly 93% of Rec. 709.

As a comparison, here the linking results with a profile of a different display that has 99% Rec. 709 coverage:

3D LUT (65^3 inverse linking): 1 minute 40 seconds
3D LUT (17^3 inverse linking): ~5 seconds

All 3D LUT tests used Rec. 709 as source gamut in conjunction with BT.1886 and abs. colorimetric + whitepoint scaling.

I think the speed is fine.

DCG (dispcalGUI) - Graphical front-end for Argyll CMS display calibration and characterization
Current stable version 2.6
Current development snapshot 2.9.1.3 beta (2015-04-29) - via 0install or standalone - latest changes

Last edited by fhoech; Yesterday at 10:37 PM.
fhoech is offline  
post #708 of 709 Unread Today, 02:54 AM
Senior Member
 
anta1974's Avatar
 
Join Date: Jun 2009
Location: Ukraine
Posts: 344
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 20 Post(s)
Liked: 22
Where i look version argyllCMS?
1.7b6 ?
Where write b6?
anta1974 is online now  
post #709 of 709 Unread Today, 03:23 AM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 5,508
Mentioned: 61 Post(s)
Tagged: 0 Thread(s)
Quoted: 748 Post(s)
Liked: 644
Quote:
Originally Posted by gwgill View Post
But is the speed of profiling & link/3DLut creation very important in the end ? Most of the time will be spent waiting for sufficient measurements to be made, and you would hope you don't have to do it very often.
No, ultimately it's the results that matter most and adding a few minutes to what is typically a 2 hour process is not important. And so far ArgyllCMS has been able to handle anything I've thrown at it including some very oddball display settings so it is a very robust approach.
zoyd is online now  
Sponsored Links
Advertisement
 
Reply Display Calibration



Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off