3D LUT Calibration - Page 4

I'm not sure if my display does crayon. It must be here in the Control Panel somewhere. Anyway, you'd probably scribble it in LUTSpeak, a language with which I'm only beginning to become familiar.

I'm not particularly defending anything, merely trying to learn through questions.

Unless its completely trashed an uncalibrated display or a display that has been setup to maximise its native precision is going to have performance characteristics that fall within the realm of displaying a video image ( not log , not linear , not DCi). It may have grayscale inaccuracy , it may have non-linearity issues however its still going to perform within a envelope that can be categorised as video rather than any other baseline. I can tell this by looking at the resultant images whether you believe me or not is immaterial although I find it hard to grasp why someone with a background in color management finds this difficult to believe.

I can tell a few things by looking at the marcie in those images , namely ...the end target is notionally a video display : its not log or linear and there isn't really much left after that in terms of options.

The white ref target is towards the warmer end of the range hence the reason for marcie looking overly cold on a properly calibrated rec.709 monitor , so the display is either closer to D.55 than D.65 or the film to video rebalance is not being applied by the lut.



I beg to differ for the reasons I've already outlined.


The 1dlut version looks as if its had the D.55 white ref for print rebalanced to D.65 , the 3dlut version doesn't (quite obviously so). Surely its easy enough for you to tell us what the luts are doing? You could start with the end targets.



In those marcie images the most dramatic visual difference is down to targeting a print gamut with the 3dlut for the log to video conversion. This is not indicative of the visual difference that would likely result from correcting the gamut on a rec.709 type display for rec.709 material. Once again I ask you how many people on this forum are going to be grading 10bit log material into video? Stick to relevance instead of sensationalist examples.


I never mentioned the prices of cinespace or truelight however I have mentioned them previously in the past and I've also mentioned the fact that there are 3dlut profile based CMS that other people on here have created for free. Yes they have limitations however they are obviously less limited than your currently non-existing device.


I never said you did , I'm re-quoting the article in question.


Accuracy is a function of hardware tolerances and measurement accuracy.
I see little point in discussing how notionally accurate a non-existent 3dlut based box would be if the sensor equipment required to attain that level of accuracy is beyond the reach of 99.9% of the user base. What is the least expensive useful sensor that you envisage your system producing meaningful results with?

Whilst I totally defend the merits of an automated profile and 3dlut calibration system ( and have done for many years) I do not see something as simple as a rec.709 gamut as being impossible to achieve adequately with a conventional CMS. At least from the perspective of home cinema display , I would imagine the differences a 3dlut system would make for all but the most seriously gamut skewed displays , would be very very subtle indeed to the point of being a non-issue without rigorous ABing within narrow range of image content.


We seem to have come back round to the hdlink pro which has been available for years , i assume by "pro market" you have discovered the problems associated with hdmi/hdcp implementation.


Just a hunch.


Nope the whole second coming/reinvention of the wheel type vibe at the mere mention of 3dluts coupled with willful ignorance of real world practicalities is what I expect of just about every color management consultant I've ever met .

If you bring this to market in usable , practical form then I will happily be the first to sing your praises. However until that time you may as well wax lyrical as to how it would be much better to use a boeing 737 to do the weekly shop rather than a family saloon because it has more cargo space.

Really? I think this one is going just swimmingly.

I use 2 3dlut files generated by yCMS used in conjunction with MadVR video render for my HTPC. I haven't been able to meter the results yet but so far I can notice an improvement in depth with my Panny S2. (the S2 suffers from bad gamma/luminance, last reading I did it came in at 1.76) The yCMS system helps to correct gamma issues and get closer to 2.22 or whatever gamma you specify.

I plan to do a measurement when I have some time, and hope to see an improvement in both the CIE, luminance, and gamma charts.

If this works, I may have to pick up a blu-ray drive and just play my blu-rays through my HTPC.

Update: I went ahead and measured the signal from my HTPC to my Panny S2 and indeed it improved my gamma from an average of 1.76 to 2.14

I took a bit of a hit in de's for the grayscale but the improved gamma is probably worth it in the long run. Here are some charts and how things are looking now.



Uploaded with ImageShack.us

And this is the CIE chart



Uploaded with ImageShack.us

And finally my RGB, the dE's got a bit worse, but I think ultimately the increased gamma and color's should make the picture look better overall. (my de's went from 0.3 to 2.7 from 30-100 IRE and now with yCMS they are 0.6 to 4.1)



Uploaded with ImageShack.us

Can you share some technical details of the ColorBox? Which internal bitdepth does it calculate in? Does it support DeepColor output? Does it dither the output for 8bit displays? Which 3dlut sizes will it support? Which interpolation will it use (trilinear, tricubic etc)? And finally, is there any information about expected prices available already?

Thanks.

http://www.upsilonsoftware.com/

Noticed John has moved the Upsilon Mixer out of Beta into retail.

It supports pretty much every popular sensor out there.

I use it for WMC LiveTV and will be moving from TMT to MPC-HC with MadVR for BD. That should represent pretty much the high water mark of consumer level playback be it PC or standalone.

It has custom gamma , black point and white point mapping and will automatically switch lut application depending on format.

I appraised it before and after with Calman and I get the most accurate end gamut correction I've ever seen regardless of price.

That reminds me, I gotta notify JohnAd about another change in the 3dlut requirements of the latest madVR version...

Yes, thanks for your update

I'll try and get the lut generator working with the latest version of madvr later on in the week.

John

Thanks!

Its possible the decrease in deltaE is down to precision limitations on the display ( the 3dlut correction may have kicked up some slight banding).

Generally I advise people to maximise the precision on the display. This entails letting the white point scale to its maximum reachable value on the hardware without any of the color channels clipping. Its sometimes not that useful to calibrate the display hardware itself in terms of gamma or grayscale as the control mechanisms on the display usually have less precision than letting the graphics card handle it. On digital displays (unless they are very high end) I never target a reference white point luminance value , just let the display hit as high as it can without clipping. Maximising precision generally is more visually important than how bright white is. You can always dial it back if its burning your eyeballs.

All it takes is a little bit of rounding error to happen in an intensity range you are metering to skew your deltaE. In your case its only around 4 still and I suspect the tolerance of your meter means its somewhat redundant to try to improve it ( or at least worry about it) unless it has an obvious visual impact ( which I doubt).


I get a very similar result when using the upsilon mixer on my totally gamut trashed panasonic plasma , one region in the whites ( think its about 80) creeps up to about 4 deltaE from 3 deltaE without the lutting. However with the lutting 60 improves from 3 to 1ish which is far more visually impactful than the decrease in accuracy at 80 (can't see it myself). The resulting gamut figures are bang on and again coupled with the fact I'm using a spyder3 of limited accuracy I suspect quantify improvement down to 1 or 2 deltaE is firmly in the realm of slicing the onion too finely.

We also need to take into account that yCMS is not aiming for best measurement results. yCMS is applying chromatic adaptation. Which means that the white point is not corrected at all. Instead all other colors are changed to match the display's native white point. Due to how our eyes/brain work the subjective image quality should be just fine, but measurements can suffer, which can increase the deltaE values. On the positive side, doing it this way, yCMS calibration doesn't lose peak brightness. The usual calibration method of correcting the white point means you usually lose peak brightness.

Well, I'm not an expert in these things, so some of what I wrote above may be incorrect. But that's how I understood it, when yesgrey explained yCMS' math to me.

Mr.D,

Apparently, people don't like it when you tell them the emperor had no clothes.

Yes I'm used to seeing a startling number of regal naturists these days. I'm sure there are more of them now than ever.

Looking forward to messing about with this. Just wish I could get a break from making films and have the time to watch some of them.

Version 1.0.2 now works with the latest madVR.

John

Gotta be frustrating, being in the movie business and not being able to actually watch any movies yourself!!


Great - thanks!

I've been trying to sit down and watch the criterion edition of seven samurai since last christmas....not a very good demonstration of gamut correction and chroma upsampling technique though!

Up-To-Date 3D LUT instructions for HTPC users here: How To Get Perfect Video Color

There is something in the US called "Fair Use".... that gives a lot of lattitude to an author to take snippets of copyrightted material and include them in editorial content.. Also that is on another website so your beef would be with the postor on that othersite (Interact) run by JRiver..

"Fair Use" isn't solely a US legal concept, but derives from the UK concept of "fair abridgement". The post in question contains no written material concerning the Light Illusions product and so does not appear to fit the "editorial" requirement of the Fair Use doctrine. A review of the LI product would, on the other hand, meet that requirement so the referenced graphics could be used for illustrative purposes. And the poster on the other site was "hulkss", who also made the post here that Light Illusions was responding to...
Edited by Rolls-Royce - 2/18/13 at 4:27pm

I think it is often maintained by message based website operators that A: they are not respsonsible for content posted on their sites and B the user content is eiditoral in nature as it is largely opinion. I am not a lawyer nor am I pretending to be one, least of all one versed in the ever shifting use or misues of intellectual property on the internet.

I think the post Light Illusion is objecting to is lame... but the internet is filled with endless lame content that is much more agregious than that particular post both in its lameness and its violation of intellectual property. It is certainly Litght Illusions right and probably obligation to object to any misuse of their graphic lest it become public domain.

I think the larger thrust is that right now the particular approach described in the offending post on Interact on how to achieve a useful 3D LUT on the PC is not viable. Open source and freeware based 3 LUTs for PCs are a bit of larger threat in the long run to software companies offering expensive solutions based around dedicate black boxes. I have a Radiance and use ChromaPure with it. But I stiil would asbolutely love to have a high quality community supported PC based 3D LUT app that worked correctly with madVR.

They may not be true 3D LUTs as you define them. They are the best I could do with freely available software tools and the results look great.

Your images are removed as requested.

The method I posted at JRiver to create a 3D LUT for madVR is the best I could do with the software tools that I could find. If you have any suggestions to improve it, I would like to hear them. It does produce very good results on the displays I have access to including my JVC projector.

Maybe I should give Upsilon Mixer a try?
Edited by hulkss - 2/18/13 at 10:47pm

madVR naively applies any given 3dlut. The quality of the madVR 3dlut processing solely depends on the 3dlut files created by external tools. So what we really need is a good software to create "perfect" 3dlut files in a suitable format for madVR. yCMS is not bad, but it's not delivering the full power of what a good 3dlut could potentially do.

Maybe I should talk to SpectraCal to make their "AutoCube" 3dlut creation solution work with madVR. @Steve / Light Illusion, if you have interest in making your software work with madVR, please feel free to contact me at madshi (at) gmail (dot) com. I've already prepared some code to make madVR work as a remotely controllable test pattern generator a long time ago. Just didn't have a chance to finish it yet. FWIW, madVR currently supports 6-8bit input 3dluts (64x64x64, 128x128x128 or 256x256x256) which always output 16bit. I know that madVR's 3dluts are bigger than is usual used. It might require a different way to calculate the 3dluts (interpolate the measurements up instead of down), but it should allow both gamma and gamut correction in one step. At least that was the original goal. If needed, of course I can make madVR support smaller 3dluts, too. But if they get much smaller, I guess I may need to do tricubic interpolation to keep quality as high as possible, which would slow GPU processing down. Currently I'm using trilinear interpolation, only, which is extremely cheap (almost free) on modern GPUs, and which is more than good enough with the really big 3dluts madVR currently uses.

Hi Steve,

In what way would the extra accuracy you refer to be of visual value?

Is it just of academic interest or could our far from perfect human eyes detect the difference?