Is there any recommended alternate whitepoints for PFS phospor monitors?

Home Forums General Discussion Is there any recommended alternate whitepoints for PFS phospor monitors?

Viewing 15 posts - 76 through 90 (of 93 total)
  • Author
    Posts
  • #25153

    AstralStorm
    Participant
    • Offline

    What HW limitations?

    I said *profile* not calibration. The profile is being applied by mpv using a 256x256x256 cube, generated by lcms2 and the result is applied in rgba32f with linear interpolation. And it’s banded at high saturations only.

    Same result when applying the 3dlut with Reshade in rgb10_a2.

    The profile also looks weird, with easily visible steps in B2A1 curve. Even the good one has a bit of a step response there, only if I cut to something close to sRGB gamut I get no such artifact.

    • This reply was modified 3 years, 9 months ago by AstralStorm.
    • This reply was modified 3 years, 9 months ago by AstralStorm.
    #25156

    AstralStorm
    Participant
    • Offline

    Well, not just steps but also discontinuities. The better profile has much less of it. Using huge amount of points in LUT also helps with this issue. It’s only B2A1 output that has this issue, other curves look as expected.
    It’s as if Argyll is trying to reject a piece of high brightness content but failing, or runs into an internal resolution problem.

    • This reply was modified 3 years, 9 months ago by AstralStorm.
    • This reply was modified 3 years, 9 months ago by AstralStorm.
    • This reply was modified 3 years, 9 months ago by AstralStorm.
    #25160

    AstralStorm
    Participant
    • Offline

    > Same result when applying the 3dlut with Reshade in rgb10_a2.

    Actually not if I pick smaller LUT size. The optimizer seems to smooth the banding problem so it’s not an issue, while accuracy is preserved.

    #25161

    Vincent
    Participant
    • Offline

    With cal there are HW limitations: limited LUT bitdepth and lack of dithering. If you said there is linear calibration and all conversions done by some color management engine, my previous claims do not hold, but as in other apps (GIMP, PS w/o 10bit & etc) that do not dither output the own engine and truncation to GPU output may be the cause. IDNK if mpv has such features (dithered output) but I remember that some openGL output can do that in mpv or did it in the past. Not an expert in mpv, but if try it if you didn’t.
    Also if you are a Windows user, try a player that supports some kind of dither, just to be sure that color management engine + posible undithered truncation to GOU causes your issues. madVR should work, for example.

    Regarding your last message, long time ago table/lut profiles caused huge ripples in some apps with Argyll profiles, not sure if it is what you mean with discontinuities. Displaycal added some smoothing somewheer but it was long time ago and I do not remember what was the solution.
    Report it as a bug with measurement data, maybe Mr. Gill will find what’s happeing… which IMHO seems to be not related at all to your claims about observer.

    #25162

    AstralStorm
    Participant
    • Offline

    There is a possibility I could try here, since the problem is at high saturations only really, I could make a test chart with many points in high saturation and high brightness ranges.

    There are also much better results if I force the preconditioning profile to be DCI-P3 or DisplayP3. So it might be a sampling related problem actually… Compounded with a probe problem.

    #25163

    AstralStorm
    Participant
    • Offline

    With cal there are HW limitations: limited LUT bitdepth and lack of dithering. If you said there is linear calibration and all conversions done by some color management engine, my previous claims do not hold, but as in other apps (GIMP, PS w/o 10bit & etc) that do not dither output the own engine and truncation to GPU output may be the cause. IDNK if mpv has such features (dithered output) but I remember that some openGL output can do that in mpv or did it in the past. Not an expert in mpv, but if try it if you didn’t.

    mpv outputs to rgb10_a2 surface (best windows compositor can handle, my choice) dithered from rgba32f in final steps with a pseudorandom dither (called fruit). Can also output rgba8 with 8-bit dither, which does not affect anything one bit. Other formats are at the mercy of GPU of course, but AMD Radeons seem to apply a dither if needed.

    The banding is much too strong and specifically located to be just dither-related though.

    Also if you are a Windows user, try a player that supports some kind of dither, just to be sure that color management engine + posible undithered truncation to GOU causes your issues. madVR should work, for example.

    Same issue with madVR if 64^3 size 3dlut is used. (Still less than if mpv uses the whole 256^3 cube.)

    Regarding your last message, long time ago table/lut profiles caused huge ripples in some apps with Argyll profiles, not sure if it is what you mean with discontinuities. Displaycal added some smoothing somewheer but it was long time ago and I do not remember what was the solution.
    Report it as a bug with measurement data, maybe Mr. Gill will find what’s happeing… which IMHO seems to be not related at all to your claims about observer.

    Indeed. Observer and correction matrix might be only incidentally related with numerical reasons or because a different set of patches is generated.

    I’m trying with a special fixed point testchart now to see how that affects the generated profile.

    #25180

    AstralStorm
    Participant
    • Offline

    Here’s the least banded profile, it’s generally monitor default settings. Mild banding at darkest blacks is visible. Note one way gamut is almost fine (with weirdness on the maximum saturation), the other way is complete trash.

    XV273K gamma 2.6 max brightness black boost 5 profile 7z

    If I touch any contrast control, I get (much more visible) banding at max saturations in addition to low brightness. No idea what’s going on here. i1 Studio does *slightly* better on it, but it’s again probably incidental.

    Calibrite ColorChecker Studio on Amazon  
    Disclosure: As an Amazon Associate I earn from qualifying purchases.

    #25181

    Vincent
    Participant
    • Offline

    Since Contrast is limiting panel response, most of them in a digital way, it looks like poor control electronics, undithered or poorly dithered.

    Did you try

    -to set display to factory contrast,

    -get your desired white visually or numerically, then measure it (write coordinates your the measurement device of you choosing),

    -set display to native white (factory gain/offset, usually “MAX CONTRAST” setting, to minimize weird behavior of control electronics)

    -set desired brightness

    -make a synth profile Rec709 to your “desired”/”alt” whitepoint coordinates (the ones you wrote before)

    profile display XYZLUT no idealization, then make a LUT3D (abs col ) for madVR with alternative Rec709 profile made before. Since all dither is done in GPU and it should work “smooth” and you minimize all weirdness due to control electronics in that display.

    If does not work for desktop buy if that monitor has 2 user settings on OSD you can eve have waht you have now for computer work and the other user preset for movies.

    If there is something wrong or oversimplified (like that contrast OSD responde) in monitor itself, and it looks like it is happening… IDNK other way to get rid of it rather than move to GPU all calculations and ensure dithered output to display to prevent banding.

    Maybe worth trying since it seems that you want to use that display for viewing movies or shows in the most accuarte way possible.

    • This reply was modified 3 years, 9 months ago by Vincent.
    • This reply was modified 3 years, 9 months ago by Vincent.
    #25184

    AstralStorm
    Participant
    • Offline

    This is allegedly factory contrast. Black boost is trash though.

    Setting 0 black boost, 50 contrast gets rid of banding. The display *still* is black boosted but the problem is no more. So yes, hot garbage firmware. Problem solved?

    I’m checking now if GPU controls have a problem, because I’m actually trying to match a PQ curve on this rather than gamma 2.6.
    GPU controls might be dithered and/or 10-bit.

    VCGT might also be 10 bit here in 10 bit mode and with latest Radeon drivers finally, but it’s still rather unusable due to being reset in fullscreen.

    But then actually looking at it, the result from those monitor controls is not (or not very) banded looking, just Argyll has more of a problem with it than human eye.

    • This reply was modified 3 years, 9 months ago by AstralStorm.
    #25186

    AstralStorm
    Participant
    • Offline

    I suspect there might be a faulty assumption in the optimizer for the profile that the monitor response is contiguous. It’s not in general, and especially with lower depth panels or if you sacrifice depth.

    Fixing that would require changing the optimizer to something that can handle mixed integer programming rather than linear (or polynomial) programming.

    See, looking at measurements in HCFR they’re not that bad even when quantizing.

    • This reply was modified 3 years, 9 months ago by AstralStorm.
    • This reply was modified 3 years, 9 months ago by AstralStorm.
    #25189

    AstralStorm
    Participant
    • Offline

    By not bad I mean sub 1 dE for every bit of 10 bit input available after tuning the controls. (in DCI-P3 space) So they actually match what is requested by the tool rather closely, and presumably perceptually match the color.
    The banding is not very apparent by eye without applying the profile.

    • This reply was modified 3 years, 9 months ago by AstralStorm.
    • This reply was modified 3 years, 9 months ago by AstralStorm.
    #25192

    AstralStorm
    Participant
    • Offline

    The other possibility is that FRC in this screen works differently for various colors and white, also causing issues when computing the profile, similar to as if backlight level were changing.

    #25193

    AstralStorm
    Participant
    • Offline

    Oh my, the alleged sequential polynomial programming code there is such a layer of hacks upon hacks… I’m thinking of starting with a clean slate implementation using scipy and displaycal profile writer, once I understand what all the hacks are supposed to do.
    I presume the basic approach of spline fitting can backfire badly when sharp discontinuities are present, like even minimal amounts of banding.

    #25194

    AstralStorm
    Participant
    • Offline

    So, the bug is either in gamut/nearsmth.c or rspl/rev.c. The gamut vector “smoothing” is one possible cause for such a problem (including “loops” at gamut peaks), the other is overshoot in interpolated spline for reverse lookup. It will take some time to figure out how exactly Argyll works there and wade through the layers of undocumented hacks, but ultimately this is very fixable and within my actual field of expertise even though as you saw I had major holes in actual understanding how Argyll works.
    (I do mess around with AI and Bayesian optimizers. Did some actual work in audio using these.)
    I’ll contact mr Gill (or rather argyll-dev ml) once I have something tangible.

    On the other hand, I could write a separate non-Argyll optimizer as well, in Python, as an experiment. It’ll take some doing.

    #25195

    AstralStorm
    Participant
    • Offline

    Could also be a bug in xicclu lookup or inverting code, which would be the easiest to fix.

Viewing 15 posts - 76 through 90 (of 93 total)

You must be logged in to reply to this topic.

Log in or Register

Display Calibration and Characterization powered by ArgyllCMS