or Connect
AVS › AVS Forum › Display Devices › Display Calibration › eeColor Processor - CalMAN - 3D LUTs
New Posts  All Forums:Forum Nav:

# eeColor Processor - CalMAN - 3D LUTs - Page 6

I think you may mean a 6^3 cube vs. 9^3

### AVS Top Picks

No I am reporting what Calman tells me.

btw, it looks like you doubled posted that. Your posts 150 and 151 are the same.
Quote:
Originally Posted by DigitalFilm

I think you may mean a 6^3 cube vs. 9^3

Sillysally, CalMAN's Method uses 9x9x9 (9-Point Cube = 729 Color Points) or 6x6x6 (6-Point Cube = 216 Color Points) using the infrormation of 16-Step WRGBCMY Luminance Measurements to calculate their interpolations.

16x9 as CalMan names is 729 Color Points + 49 Extra Luminance Points = 778 Total Color Points.

16x6 as CalMan names is 216 Color Points + 70 extra Luminance Points = 286 Total Color Points.
Quote:
Originally Posted by ConnecTEDDD

Sillysally, CalMAN's Method uses 9x9x9 (9-Point Cube = 729 Color Points) or 6x6x6 (6-Point Cube = 216 Color Points) using the infrormation of 16-Step WRGBCMY Luminance Measurements to calculate their interpolations.

16x9 as CalMan names is 729 Color Points + 49 Extra Luminance Points = 778 Total Color Points.

16x6 as CalMan names is 216 Color Points + 70 extra Luminance Points = 286 Total Color Points.

Yes you are correct, however because this thread is for Calman and EE I am just trying to keep it simple and use the terms Calman uses when setting up a LUT/cube calibration.

ss
Quote:
Originally Posted by sillysally

Yes you are correct, however because this thread is for Calman and EE I am just trying to keep it simple and use the terms Calman uses when setting up a LUT/cube calibration.

ss

We do list our cube calibrations by luminance x cube size. So yes the first number of a 16x9 is the number of WRGBCMY points we calibrate first to turn the cube into a linear space. The second number in the 16x9 is the cube size so a 9^3. This gives us the flexibility to do luminance independent of saturations. But also allows for a full17^3 if needed or chosen.
Quote:
Originally Posted by derekjsmith

But also allows for a full17^3 if needed or chosen.

Sillysally, do you have time to run a true 17-Point Cube (4913 Color Points) with CalMAN to tell us how it performs?
Edited by ConnecTEDDD - 3/26/13 at 4:41pm
Quote:
Originally Posted by ConnecTEDDD

SilliSally, do you have time to run a true 17-Point Cube (4913 Color Points) with CalMAN to tell us how it performs?

I would if I could, but as I said I can't get a 778/1 points to run right.
I did slim down the programs running in my LT (laptop) and got a new RS 232 (for the Mini) and USB 2.0 A-Male to B-Male Cable (for the eecolor). So maybe I will give a 778/1 points a try tonight and see if I can get that far first. Then wait for the updated beta from Calman to come out.
I am happy with my 278 point cube, I am using now.

Has it been confirmed that eecolor can't pass through any lossless MC audio?. It makes no never mind to me because I only run video out of HDMI port 1 from my BDP-95, HDMI port 2 is for my audio. .

ss

Update

Ran 771 point lut cube, took about 3 1/2 hours 2250 reads, Had NP this time all went very smoothly. Charts look good.
So I guess what ever I did worked and Calman was not to blame. Now I will view some source material.

Edited by sillysally - 3/27/13 at 7:32am
Ok, so 3.5 hours for 771 patches.

A 17^3 cube profile would be 4913 patches, and based on your timings above I make that about 22 hours...

Eeek!
Quote:
Originally Posted by DigitalFilm

Ok, so 3.5 hours for 771 patches.

A 17^3 cube profile would be 4913 patches, and based on your timings above I make that about 22 hours...

Eeek!

Typically the more points we calibrate the lower the number of times we need to adjust each point. So for 771 points he need 2200 reads or 2.8 reads per point, going to 4096 points the points per read may likely drop to 2 or less. So I would doubt that it would take any longer than about 3x as long. Granted that still clocks in at 9-10hours.
I'm sorry, but that makes no sense to me at all...
You are less accurate with more points???
I think what Joel is saying that since the correction points are closer together fewer adjustments are required therefor each point don't take as long as if it was fewer points that need correcting.

just a quick question.

Does CalMAN Enthusiast support 17-Point Cube (4913 Colour Points) in the current beta? if so how does it preform? I know there is a debate if its required with this many points, but just curious
Quote:
Originally Posted by sotti

in the catalyst control center under the display settings there should be a setting for pixel format.

If you select rgb full range the output should be reasonable close.

We've put a couple ATI cards on the scope and their not perfect, but only ever off by a bit or so.

Of course a sanity check like you ran above is always a good idea.

Hi Sotti,

I have been trying to set it up correctly, but not happy with the result at all, could I get you to confirm the following:

1) Shut town CalPC Client
2) Reset any profiles with calibration tester (from x-rite) I have done this on both display's (main laptop and extended TV)
3) Set the following in CCC:

that's it?

I have not found a way to get accurate results, when watching material this way it look like the display is on dynamic mode. any suggestions?
Quote:
Originally Posted by DigitalFilm

I'm sorry, but that makes no sense to me at all...
You are less accurate with more points???

Continuing to sow the seeds of doubt I see. If I had a suspicious mind I might think that your trolling of this and the ArgyllCMS thread had a motive other than honest discussion.

- Joined this month with 16 total posts
- concern trolling CalMAN and ArgyllCMS
- all praise in the LightSpace thread

You aren't related to steve are you?
Not related, no - but I am genuinely interested in how having more points needs less accuracy in the final point calibration.

A valid question I would suggest?

As for LightSpace thread, I agree with the granularity concept, and it seems the tests Buzz has done also suggest there is something in that.

Makes sense to me. Does it not to you?

And the reason for the threads posted in is my interest in in 3D LUT calibration only - not in other approaches.

There is some new 3D LUT hardware coming to market, and that's my real interest.
Edited by DigitalFilm - 3/29/13 at 7:58am
Quote:
Originally Posted by DigitalFilm

Not related, no - but I am genuinely interested in how having more points needs less accuracy in the final point calibration.

A valid question I would suggest?

As for LightSpace thread, I agree with the granularity concept, and it seems the tests Buzz has done also suggest there is something in that.

Makes sense to me. Does it not to you?

calman work the following way

then correct
re measure/validate

if the validate is ok skip to next color point if not then correct an revalidate.

since the points is closer to each other when you have many color points its only natural that more points are either validated with first read or only require correction once.
@Visca Blaugrana

Not sure it would work if you insist on using CCC but try unchecking 'Use my settings for this device' in the 'Devices' tab of Microsofts 'Color Management' in the Control Panel.

Strangely, any newly calibrated (ie profiled) display has its profile hijacked and used by my laptop thus defeating the Xrite Tool unless I do this.
Quote:
since the points is closer to each other when you have many color points its only natural that more points are either validated with first read or only require correction once.
I'm not sure that is correct?

If a given colour point requires 3 or 4 reads/changes to make accurate when using a given number of points, increasing the number of points wont make the original colour more accurate?

Or am I missing something?
I would love not using ccc, but do I not need it to set the correct output?

am I better of.

stop the ccc and MOM service.
remove all profiles in color management
and de select as you suggest above?
and of course still use calibration tester
Quote:
Originally Posted by visca blaugrana

that's it?

I have not found a way to get accurate results, when watching material this way it look like the display is on dynamic mode. any suggestions?

You probably want to check the dynamic range and force video content t stay at 16-235. The rest of the setting look correct.

For 99.9% of users they want their video content expanded to match the desktop.
Quote:
Originally Posted by visca blaugrana

I would love not using ccc, but do I not need it to set the correct output?

am I better of.

stop the ccc and MOM service.
remove all profiles in color management
and de select as you suggest above?
and of course still use calibration tester

My interest is in using my graphics card to generate accurate patterns because my DUO, despite being a magnificent Pattern Generator, cannot integrate with the triplet method preferred by Calman. (DUO uses 0-100%).

The Graphics Card solution is a no cost method so I have simply used the Xrite tool to try to stop any interference by existing profiles.

In Color Management I do not have any Profile selected yet Calman 5.1 gives me the patterns I need.

I have checked these patterns against AVSHD generated ones and the difference is negligible. Others claim this is not the case.
Edited by PE06MCG - 3/29/13 at 8:35am
Quote:
Originally Posted by DigitalFilm

Not related, no - but I am genuinely interested in how having more points needs less accuracy in the final point calibration.

A valid question I would suggest?

In a statistical problem like this you can decrease the precision on each measurement as you increase the number of measurements and still arrive at the same mean error. There is nothing special about that.

Quote:
As for LightSpace thread, I agree with the granularity concept, and it seems the tests Buzz has done also suggest there is something in that.

Makes sense to me. Does it not to you?

Of course it makes sense, but higher inefficient granularity is just dumb. You can achieve the same mean error using 17^3 grid points as using 1500 points optimized using the geodesic algorithms present in ArgyllCMS.
Quote:
And the reason for the threads posted in is my interest in in 3D LUT calibration only - not in other approaches.

There is some new 3D LUT hardware coming to market, and that's my real interest.

ArgyllCMS is a 3DLUT approach, equivalent in it's math to both CalMAN and LightSpace, performs just as well, has the additional capability of doing CIECAM02 appearance matching, and it's free. The .icm profiles it uses to link source and target color spaces are just a data path for exactly the same XYZ measurements and manipulations that LightSpace uses.
Edited by zoyd - 3/29/13 at 9:17am
Quote:
Originally Posted by DigitalFilm

I'm not sure that is correct?

If a given colour point requires 3 or 4 reads/changes to make accurate when using a given number of points, increasing the number of points wont make the original colour more accurate?

Or am I missing something?

Yes you are missing something.

Think of just a grayscale ramp.

At first if it's widely uncalibrated adjusting any point would take a large number or readings.

But if you calibrate a 10 point grayscale first and then look at a 21 point ramp, the 5% steps are going to be much more accurate after the 10 point adjustment.

That's the basic premise of how our code works. We calibrate the points in a specific order so that with each set of points we add to the cube, the accuracy of the entire cube goes up, which means the next set of points we measure will be more accurately calibrated to start with.
Quote:
Originally Posted by sotti

Yes you are missing something.

Think of just a grayscale ramp.

At first if it's widely uncalibrated adjusting any point would take a large number or readings.

But if you calibrate a 10 point grayscale first and then look at a 21 point ramp, the 5% steps are going to be much more accurate after the 10 point adjustment.

That's the basic premise of how our code works. We calibrate the points in a specific order so that with each set of points we add to the cube, the accuracy of the entire cube goes up, which means the next set of points we measure will be more accurately calibrated to start with.

Hi Joel,

There used to be train of thought that changing values at say 20% luminance had a potential effect on say 80%. Is that still the case?.

In other words can extra calibration ruin a calibration elsewhere rather than relying on interpolation?
While I appreciate what is being said, I would think that the non-linearity of displays would mean it is just as possible for the initial change to make other points more inaccurate?

This is associated to what PE06MCG is saying above.

Not totally sure, just wondering.
Quote:
Originally Posted by DigitalFilm

While I appreciate what is being said, I would think that the non-linearity of displays would mean it is just as possible for the initial change to make other points more inaccurate?

This is associated to what PE06MCG is saying above.

Not totally sure, just wondering.

We've done extensive modeling on our algorithms as well as field testing. We don't find that to be the case.
Quote:
Originally Posted by PE06MCG

Hi Joel,

There used to be train of thought that changing values at say 20% luminance had a potential effect on say 80%. Is that still the case?.

In other words can extra calibration ruin a calibration elsewhere rather than relying on interpolation?

For a 3DLUT this is not the case. Each fixed point we adjust is completely discrete, and on the output side in hardware each input triplet is only affected by the 8 bounding LUT entries.

A Sharp Elite is a totally different animal, so it definitely depends on the design of the CMS, but for 3D LUTs, all changes remain local to their area of effect.
Quote:
Originally Posted by DigitalFilm

While I appreciate what is being said, I would think that the non-linearity of displays would mean it is just as possible for the initial change to make other points more inaccurate?

This is associated to what PE06MCG is saying above.

Not totally sure, just wondering.

The reason this doesn't happen is that measurable non-linearities occur over fairly broad device coordinates for typical displays so that to a well designed sampling set the corrections are being applied to locally smooth volumes. The trick is finding a sample space that is dense enough to be locally smooth, CalMAN appears to have done a very good job at that.
Quote:
Originally Posted by sotti

For a 3DLUT this is not the case. Each fixed point we adjust is completely discrete, and on the output side in hardware each input triplet is only affected by the 8 bounding LUT entries.

A Sharp Elite is a totally different animal, so it definitely depends on the design of the CMS, but for 3D LUTs, all changes remain local to their area of effect.

Thanks for that clarification.

I can now understand that having that predictability can make your process to calibrate 3D LUT's slicker.
Quote:
Originally Posted by sotti

The reason deep color can be help is that when the data, and all blu-rays are 8bit (no more, no less), goes into the EE, it's padded up 16bit, processed for color, and then mapped back to 8, 10, or 12bit for output. So the data on output can actually be >8bit with actual information int the final few bits.
Quote:
Originally Posted by sotti

Not necessarily.

It's storing the data as 10 bit to fit it in the flash memory. But when it selects a LUT, it could easily be padding that data and processing the internal signal at a higher bit depth.

regarding the bit-depth of processing and output, here's direct info from eeColor:

> 3D LUT algorithm only supports maximum of 10 bits input so the data will be reduced from 12 bits to 10 bits if needed.

> max bit-depth of internal processing: 10 Bits for the 3D LUT.

> max resolution: 1080P is the offical max supported resolution.

> 12 bits HDMI in will be 12 bits HDMI out but will be rescaled to 10 bits for 3D color table processing.

- M
Thanks for the verification.

I was doing pure speculation on what it could possibly be doing with the information i had available. Good to have confirmation.
New Posts  All Forums:Forum Nav:
Return Home
Back to Forum: Display Calibration
• eeColor Processor - CalMAN - 3D LUTs

### AVS Top Picks

AVS › AVS Forum › Display Devices › Display Calibration › eeColor Processor - CalMAN - 3D LUTs