- This topic has 6 replies, 2 voices, and was last updated 5 years, 1 month ago by
.
-
Posts
-
Hi,
I have Firefox configured correctly to load ICC profiles. E.g. gfx.color_management.enablev4=true, gfx.color_management.mode=1. Until recently I have been using a profile created by someone else for the matte version of my screen using a Spyder 3 (I have the glossy), and the result is actually quite decent. But I’ve been wanting better results, so when I was recently able to get a Spyder 3 of my own I made a profile. It’s installed with colord (under KDE on Linux) and looks great in every (color managed) program I’ve tried – except Firefox. In Firefox color management has stopped working altogether – colors in the GUI are washed out and Firefox seems to be falling back to assuming the display profile is sRGB – so images loaded in the browser are over-saturated.
I have checked that Firefox goes back to working correctly if I manually specify the profile I used to use with gfx.color_management.display_profile. Switching to a swapped color profile also gives the expected results in Firefox, so nothing appears to be unusual on that front. Something is going wrong with the DisplayCAL created profile.
I’ve attached the profile. Is there anything I can do to debug this?
Thanks!
displaycal 3.7.2.0; argyllcms 2.0.1; colord 1.4.4; firefox 66.0.1
Hi,
the profile itself shows that the Spyder3 is struggling to measure the near-black response – which is not at all unexpected, it really hasn’t much low light capability to speak of, which is also exacerbated by the ambient adjustment you enabled, which pushes the display gamma to a hefty 3.0 value.
My recommendation:
- Disable calibration ambient light adjustment
- Reduce the number of profiling patches to the default 175 (testchart “Auto”, move the slider), there really is no benefit to be had with measuring more patches with a Spyder3
- Re-calibrate and profile
I’ve done this; I still have the same problem in Firefox. The new profile is attached.
Since the profiles I’m creating with DisplayCAL work and look just fine in every application other than Firefox, I think it’s fairly safe to say the problem is not with the calibration? In addition, the new profile I just made looks significantly worse to my eye than the old one (above). The output of the VCGT curves is much too bright, and to my eye slightly too red, compared to the old profile. I’m not sure why that is, though.
Anyway, the main issue for me is just getting Firefox to accept the profile. I’m happy enough with the calibration result as is.
One question: since you say there’s no point in using > 175 patches, would you say it’s useful to put calibration speed on “low”, as I did for the first profile?
Thanks for your help!
That profile looks much better. Self check is sane (avg < 0.12 dE76, peak 1.88), near-black response is smooth, overall display response after calibration is gamma 2.2.
The output of the VCGT curves is much too bright,
How bright is your viewing environment? Monitor brightness of 200 cd/m2 can be a lot if you’re in a dim room. If your Spyder3 has the ambient sensor, you can use it to set a luminance target for calibration (on the calibration tab, set white level to custom and click the “measure ambient” button). Otherwise, adjust white brightness to a comfortable level before calibration.
and to my eye slightly too red
You can set the usual 6500K whitepoint target if needed.
Since the profiles I’m creating with DisplayCAL work and look just fine in every application other than Firefox, I think it’s fairly safe to say the problem is not with the calibration?
How do you verify that Firefox (or any other application for that matter) is using the profile?
One question: since you say there’s no point in using > 175 patches
With a Spyder3 there’s no point, and on monitors with good linearity.
would you say it’s useful to put calibration speed on “low”, as I did for the first profile?
No. A lot of time will be spent measuring the dark region of the gray axis, one of the Spyder3’s weak points. Use the default of “fast”
That profile looks much better. Self check is sane (avg < 0.12 dE76, peak 1.88), near-black response is smooth, overall display response after calibration is gamma 2.2.
I had a better profile with similar self-check values that I made earlier, so if that’s actually important I’ll switch to that one. The only reason I didn’t use it was that it was made with “very fast” settings, but given what you’ve said about the Spyder 3 it might not matter.
You can set the usual 6500K whitepoint target if needed.
I did this at first, but every profile I created with the 6500k point set came out bright yellow green. This is a laptop screen and doesn’t have an adjustable white point, and for some reason I’m not getting a decent calibration if I try to force it to 6500k. Accepting the native white point has been giving me much better results.
How do you verify that Firefox (or any other application for that matter) is using the profile?
By visual inspection. Is there another way? As I said in my first post, Firefox looks correct (i.e. the same as every other color managed app) when I use any other profile. That includes color swap profiles which make it extremely obvious that the profile is being used. My process for these DisplayCAL profiles is exactly the same as with the other profiles. I set the calibration profile in colord, and either manually specify the profile in Firefox’s about:config or leave it unset (it make no difference, Firefox is picking up the profile from colord correctly).
I then compare the results between color managed applications and with two other profiled devices I have. There is perfect agreement between every piece of color managed software (GIMP, Gnome’s image viewer, etc), and near-perfect agreement with other devices like a Macbook with a retina screen. The results in Firefox are absurd and obviously wrong. Images are greatly over-saturated, which suggests that Firefox is rejecting the profile and using sRGB instead. Other images like favicons and GUI icons are extremely de-saturated, though I don’t know the technical reason Firefox might be doing that.
If you know of a way to inspect a window / image and determine how it has been color-corrected, I’d be happy to use more technical methods.
Just as a quick illustration, here is what the HTTPS lock icon looks like in Firefox with the DisplayCAL profile. The “before” is the previous Spyder 3 profile provided by someone else. These profiles look very similar in every other color managed application.
Note how the latter image is completely desaturated. The whole Firefox GUI is affected in this way. Here’s a screenshot that shows the difference in how an image is rendered in Firefox vs EOG. The latter looks exactly like the image does in GIMP and every other color managed application, and matches my Macbook very closely as well.
I did this at first, but every profile I created with the 6500k point set came out bright yellow green.
Probably time to ditch the Spyder3.
Just as a quick illustration, here is what the HTTPS lock icon looks like in Firefox with the DisplayCAL profile. The “before” is the previous Spyder 3 profile provided by someone else. These profiles look very similar in every other color managed application.
I’m pretty sure you have a profile mismatch somewhere, i.e. applications are using different display profiles, probably due to the several mechanisms to query a profile not being in sync. E.g. Firefox AFAIK is not colord-aware, I’m not sure about EOG, https://www.freedesktop.org/software/colord/faq.html#application-support seems to indicate it doesn’t use colord as well, but I am not sure if that information is current.
Have you run the DisplayCAL profile loader (displaycal-apply-profiles)? That might be able to sync everything up (it tries to query colord if available and then sets the _ICC_PROFILE atoms accordingly).
If you know of a way to inspect a window / image and determine how it has been color-corrected, I’d be happy to use more technical methods.
You could install one of the “Test” profiles that come with DisplayCAL (/usr/share/DisplayCAL/tests, choose one of the _clut variants) which will introduce a strong color cast via the calibration curves. The profile itself will cancel out the color cast when used, so it should be easy to determine which color managed app uses the profile.
