GCD - Gamut Calibration Disk - Page 19

Hello zoyd,
Thanks for the clarification. Now I hope Sotti and Tom will chip in to explain how their software does it and then to achieve a common standard.
To me the correct way seems to be to use bit level stimulus for the calculation

Here is another mathematical way to represent the same sort of idea, where rounding to the nearest 8-bit luma value puts an error in the middle of the grayscale, and rounding-down places any discrepancy near black. The idea of the following spreadsheet is to relate the "Actual Percent" with the "Target Luminance" and look at how it would affect display gamma. I used http://www.avsforum.com/t/843411/how-to-calculate-gamma#post_10469695 to calculate the "Point Gamma" columns and compared the point gammas using the "Gamma Change" column. I'm not really sure if the trend toward a lower gamma near black on the round-down video is explicitly an issue in relation to question 6 at http://www.poynton.com/PDFs/GammaFAQ.pdf

Encode.xls 29k .xls file

Without getting into the math, the message I take from this is that patterns need to match what the calibration software is expecting? I'll have to reread this thread more carefully to quantify the impact in terms of dE.
Perhaps it is more accurate to let calibration software output its own patterns via a laptop with HDMI. At least then there should not be any mismatch.

Perhaps only semi-appropriate here, but where does this leave people with video processors that can output their own patterns? How do we verify whether patterns match now:(

that's what I'm thinking too. Just use the levels the software expects and everything should be fine. Guess that all calculations the software does is based on what is expected and not on what's mathematical correct. See:
http://www.avsforum.com/t/1441407/grayscale-variances-different-patterns-plasma-calibration/120#post_22687374

what calman expects can easy be seen by turning on the "pattern change prompts" option.


As I'm also working on a pattern disc (5% and 6.5% windows) for testing my VT50:
how is the process from Video to 4:2:0 YCbCr and what authoring software do You use?

thanks!

no, just the opposite since the patterns are limited to 8 bit integers and the software is not. From the above gamma error calculations you can see that at most we are talking about 1-1.5% errors in gamma (do you care if your real gamma is 2.2 or 2.22 at 10% stimulus and correct at all others?) While this error is measurable it will not be noticeable to the user, so I don't expect anyone changing their software and/or patterns to be completely consistent.

You are conflating two different issues, let me try and separate them for you.

1. The discussion above concerns the values used to generate the pattern's luma value (loosely, it's intensity). Because you can't have 10 (or 20) equal integer intervals to cover the 219 possible steps you have to drop 1 step somewhere in the sequence. The only inconsistency between DVE/AVSHD and the GCD disks is where they drop that step. The end result of this choice depends on what intervals your calibration software assumes but will be no larger than ~0.02 in gamma and only for the interval with the skipped code. This is simply a non-issue.

2. The other issue has to do with what pattern size is appropriate for your display technology and also systematic errors in your own methods and equipment usage and the equipment itself. That belongs in another thread.

Basically you're measuring how your display responds differently depending on the displayed image. If you were to use an LCD-based display with a fixed backlight, you would find that DVE and AVS HD 709 are in complete agreement within the limits of the meter and the stability of the backlight. In such a situation with a contact meter, the expectation is that it doesn't matter if you measure a field, a window of various sizes, or an APL series, because the patterns all contain the same digital information in the measurement area.

Again you're talking about the display, rather than the calibration pattern.

It's been discussed before, and personally I haven't seen data that suggests choosing a particular size window increases the correlation among various displays. If anyone had provided a data-backed argument I was willing to consider changing the window size, but I never saw anyone working on data that concentrated to show if a particular size window actually tended to bring various displays more into agreement. I don't see how it would, but I also don't care to concentrate on plasma or CRT displays.

My cynical side says there's no standard because it might not be good for business if people really started concentrating on just how much observable display to display variation there can be after "calibration". My practical side says that as long as there are various displays that operate somewhat differently, the sort of accuracy people think they're achieving may not really be possible.

Pattern generators have to conform to the same 8-bit world that they attempt to calibrate. From an end-user perspective they can't achieve anything that couldn't also be done with video.

If the noob invasion is not allowed (or dilluting this thread) please do say so and I'll take it in a separate one. I understand 8-bit integers cannot represent a decimal value.

For instance: 5%: (235-16) * 5 / 100 = 10.95 or presented as 8-bit binary 0000 1011. Rounding up to 11 would be the mathematically correct thing to do. Rounded down to 10, the binary represenation changes to 0000 1010.
This is essentially what we are discussing.
How Zoyd and Alluring get to 26 and 27 as luma values then is still a bit (no pun intended) beyond my understanding (I'm talking about the Excel spreadsheet screenshots posted). Maybe someone can round this up (sorry, small joke:-)) for those still struggeling to understand (including myself).

You're almost there, add 16 (all stimulus values are relative to black, not 0) to your rounded values.

Ah thanks, I missed that. In the Excel screenshot you posted, the AVSHD error as % luminance is actually quite large from 5% to let's say 20% stimulus?

As Shakespeare said:"Tom and sotti: where areth thou?"

zoyd is right. Patterns are limited, software is not. So the software has to follow the pattern to make things correct. So far in this discussion, we did not see a response from Sotti and Tom yet.

But of course, software following patterns only makes sense when patterns are the same, and now it is not.
Is it too much to ask that zoyd, visca, alluringreality, Tom and sotti put their heads together to agree on a standard?

And don't say it is only a small error. Because to me it looks also like just a small effort to put it right.

conceptually quite simple but to remake either of these disks would take many hours of work. Every pattern using 5% through 50% luma would have to be remastered

What size are the small windows on this disc?

Never mind, I found the answer to my last question when I checked the disk. I actually found an issue unique to myself. I could only achieve hitting a peak output target with my contrast control using the smallest windows on the GCD disk. Using any other sized window (both normal and APL) on GCD or any other disk would give me an inaccurate reading. I would try to achieve a target of 35fL and I would actually end up north of 40fL. Everything improved when I moved to the other windows

I have a pn51d6500 samsung plasma and with the Cell Light set to 20 and Contrast at 95 i get 32fL once the set is calibrated. If I push the contrast to 100 then I get 35fL. Also I noticed on my set that what I thought was "pinking "to my eye was in fact that at a contrast of 95 was actually my 2pt wb Red gain being way too high. In default movie mode with a meter my Red was over 130% appearing very pink/redish in the greyscale this falsely led me to adjust the contrast to a lower setting. Now my Red gain is at 0 and it is closer to 100% for RGB balance and I have left contrast at 95 .

I noticed something odd about the white clipping pattern on the AVCHD a5.0.0 version of the disc. It has spaces for flashing bars at digital 255 and 254 even though the background of the pattern cannot exceed 254 on a disc. It should go up to 253 only, just like on the AVSHD709 disc.

I'm not sure I follow. Are you talking about grayscale when you say 100% luminance or the dynamic range tool or gamut/saturation sweeps? What do you mean by fixed luminance patterns vs. variable luminance patterns? Are you saying I should use the AVSHD709 disc over the GCD one or select the AVS disc in CalMAN 5 over the Generic Calibration Disc option?

If the background is 255, having a bar for 255 is pointless since it will never flash as there is no difference in digital levels.

If the background is 254, having bars for 255 and 254 is pointless since they will never flash.

I believe awhile back Sotti mentioned somewhere here that you cannot encode a value higher than digital 254 on a disc but you can output 255 on a dedicated pattern generator.

I've found the Gamut Calibration Disc to work fine with the CMS portion of the CalMAN 5 Enthusiast Workflow with 100% saturation, 75% luminance/stim/amplitude test patterns... identical to the patterns on the AVSHD709 disc. Are you referring to 75% saturation, 75% stim patterns instead?

check this thread out

http://www.spectracal.com/forum/viewtopic.php?f=94&t=4279&p=28820#p28813

I have almost all available disks, i have scanned each contrast pattern, all disks are using 254 level at the background with 253 and lower flashing.

That sounds right and is what I noticed on the AVSHD709 disc.

I'm not following this one. Do you mean to create 6% windows (on black) in additional to the 10% already available? Because 6% windows on 23% APL backgrounds will not measure any differently than 10% windows on 23% backgrounds. As far as gamma calibration on plasmas is concerned the existing APL patterns are optimal based on measurement that both Tom and I did last year.

You can find the video specifications if you Google the following:

6.11 Quantization level assignment (8-bit coding)

Video data: 1 through 254
Timing references: 0 and 255

I'd say it would be helpful if the next release of the GCD got rid of the 255 and 254 bars since the background is 254. Otherwise, users might trying lowering Contrast in an effort to make those last two bars appear, but they never will for the reasons covered above. Since the pattern is effectively from/based on the White Clipping Pattern on the AVS disc, it makes sense to set it up in the same way. Also, something odd is going on with the dynamic range, clipping with peak white pattern for 109%, as shown in the link to the SpectraCal forum above.