- This topic has 11 replies, 3 voices, and was last updated 2 months, 2 weeks ago by
.
-
Posts
-
How long does it take for a moderator to approve a post? I’ve had a post sitting for almost a week now.
That’s weird. I’ve never had a post awaiting moderation. Assuming the only moderator is the DisplayCAL dev, he’s not around much these days
It was my first post so maybe that’s the reason? It had a couple of image sin it as well as a icm for reference. I’ll repost it below and send the images in a follow up in case this this goes through…
“Dark/banding colors in 32bit/channel PS with “Use graphics processor” enabled”
I’ve run into a weird problem where any image when viewed as 32-bit/channel in Photoshop looks darker/banded than the same image as 8 or 16 bit but only when “use graphics processor” is enabled. On the screen capture below (also attached), both images are 32-bit but image on the left is with “Use graphics processor” on, the right with it “off”. It’s a little difficult to tell but the shadow and darker areas in the image are overly contrasty in the left image. You can also see that the gradient doesn’t have a smooth transition at the darker end. It’s not unique to any image; even regular 8 or 16 bit images converted to 32 bit exhibit this problem. This problem only shows up in Photoshop (currently 25.4 but I’ve also rolled back to a 22 version and it’s still there). Every other app I’ve tried opening the 32bit image is fine (After effects, Lightroom, resolve, irfanview, etc.). After a long process trying to determine what might be the problem I eventually was able to determine it’s somehow related to the profiles created and installed from DisplayCAL (see below).
(Image in next post…)
I’ve calibrated my 2 Dell U2415 monitors with a Spyder 5 for many years using DisplayCAL and been very happy with it (6500k, 120 luminance). Just to make sure I also created a new profile but it also exhibited the same issues. I also updated to the latest Nvidia drivers on a 3070 although the problem also shows up on any system with “use graphics processor” enabled, even an I5 with Iris integrated graphics, so I don’t think it’s hardware related. Resetting video card gamma table didn’t change the difference between the “use graphics processor” images (although the did of course look different than the normal correct gamma table ones). I eventually was able to determine that it was related to displayCAL only after trying a clean install on a new/clean system. Everything looked fine but once I installed a profile the issue popped back up. Removing the profile reset everything back to normal. I then went and calibrated just using the Spyder 5 Express software and those profiles are working fine (for now).
So…. what could be the cause of this? Is it a problem with DisplayCAL, the sensor (the Spyder Express profile works fine though), Argyll, the profile (attached), or something else? It just seems so odd that this only shows up whenever “use graphics processor” is enabled. Any help is appreciated. Is there any more info I can provide that might help figure this out?
Create a profile win DIsplayCAL single curve + matrix + black point compensation… it’s likely that Spyder software or driver ICM is the same kind of profile.
If with that simper profile does not show bands it is Photoshop’s fault not DisplayCAL. There are several issues with rounding errors in color managed software, uisng this kind of idealized profiles minimizes those issues.
A well behaved U2415 shoudl be able to be described with primaries and idealized neutral greyscale (after GPU calibration).Yes!! That fixed it. Thanks so much. I actually went with the curves+matrix first instead of the single curve since I had taken a look at the Spyder software profile in the DisplayCAL “profile info” viewer and noticed there were three different curves (rgb). Honestly, I have no idea if that’s actually what the “curves” vs “single curve” is referring to but I thought it would be worth a shot and it worked out.
Somewhat off topic… How much of a difference should I expect between using the “curves+matrix” vs “XYZ LUt +matrix” profiles? I can see from the documentation that the XYZ can be more accurate, but how much? Also, any idea why I’ve never noticed this before? It’s probably been a year or two since I’ve needed linear in PS but has something changed?
Thank you again.
Yes, that’s the difference between curves vs single curve.
Curves produce smoother results than XYZ but can’t model complex behavior, so it really comes down to: try curves first. Verify. Good? Be happy. Not good? Try XYZ. Isn’t linear supposed to look dark?
Yes!! That fixed it. Thanks so much. I actually went with the curves+matrix first instead of the single curve since I had taken a look at the Spyder software profile in the DisplayCAL “profile info” viewer and noticed there were three different curves (rgb). Honestly, I have no idea if that’s actually what the “curves” vs “single curve” is referring to but I thought it would be worth a shot and it worked out.
Somewhat off topic… How much of a difference should I expect between using the “curves+matrix” vs “XYZ LUt +matrix” profiles? I can see from the documentation that the XYZ can be more accurate, but how much? Also, any idea why I’ve never noticed this before? It’s probably been a year or two since I’ve needed linear in PS but has something changed?
Thank you again.
XYZ maps display behavior with a 3D mesh RGB <-> XYZ, then capture also “gamma” per channel, 3 independent curves (TRC)
curves + matrix. Asume that display has an ideal additive behavior, then capture also “gamma” per channel, independent. So knowing primaries & gamma you can theoretically predict all in gamut RGB <->XYZ mappings.
single curve + matrix, same as above but asume that grey is perfectly neutral, even if it is not.
+BPC = asume infinite contrast. RGB 0 = L* 0 ouput.
single curve + matrix + bpc minimizes the rounding errors that may cause a color cast on gradients. but you need that after calibration grey is actually neutral (shoudl not be an issue with medium speed or slower) and contrast is not too low (>800:1) so not black (non RGB 0) but dark colors won’t be lifted so much.
This one is the “safest” choice if display is well behaved and the typical profile configuration for HW calibration suitesThanks for the detailed explanation Vincent. It helps explain it a bit to me although I don’t fully understand all of it (my fault not yours). Still useful though.
Old Man – Thanks for the clarification on the curves/single. For linear, I need to work in the Linear RGB color space in order to get the layers to “add” correctly. To my knowledge the only way to achieve this is in 32 bit mode. Photoshop’s “Blend RGB colors using Gamma–>1.00” isn’t enough.
Not in my experience but in general I think you’re correct and I may just be using a specific work case. If I render out a scene from 3ds max and save it as a 32 bit exr with a gamma of 1.0 (linear) and open this file in PS it looks exactly like it does in the render buffer of 3ds max. Importantly though it also has full 32 bit float values in it that I need (especially values over 1). Similarly, if I have an 8 bit image image in PS and convert it to 32 bit it should look exactly the same. I understand that true linear images are darker since there’s no gamma applied but in practice isn’t this accounted for in the display? I don’t have any doubt that I’m a little in the dark here (no pun intended) and that the process I have just works for me without a full grasp of what’s going on behind the scenes.
No, it’s not accounted for in the display unless you have a profile calibrated to linear, but then everything else would be too luminous. As long as it works for you is all that matters though.
I think the typical workflow involves doing whatever you have to do in linear, but then bringing it back to a standard gamma for viewing/rendering
