- This topic has 11 replies, 2 voices, and was last updated 6 years, 1 month ago by
.
-
Posts
-
Hey i’m a little unsure what settings to use when creating a 3dlut for uploading to the TV,i see all these settings like rendering intent,source color space, input/output encoding, input/output bitdepth etc, does anyone have any tips what to use, It’s a LG C9.
What do you want to calibrate to? Rec. 709? There’s ready-made presets for it where you just need to change the 3D LUT format.
What do you want to calibrate to? Rec. 709? There’s ready-made presets for it where you just need to change the 3D LUT format.
Yeah Rec709 I’m just a little confused at all the options like what bit depth or output/input encoding to use, I get that for full RGB 0-255 should be used but what if I convert to yCbCr limited in the GPU due to banding issues with RGB, do I use 16-235/16-235 or 0-255/16-235, I also read that for the c9 3dlut should be done with 16-255 but I can’t find that option.
The DeviceControl format doesn’t allow choice of encoding, so it will all be set correctly when choosing it.
The DeviceControl format doesn’t allow choice of encoding, so it will all be set correctly when choosing it.
Oh I thought LG used the Autodesk format according to the lg c8 thread here.
The two are similar, but not the same. Autodesk has different column ordering so can’t be used with DeviceControl, for example.
The two are similar, but not the same. Autodesk has different column ordering so can’t be used with DeviceControl, for example.
So i tried doing a profiling with both auto optimized, and the very large lut testchart, why is the results with auto optimized so much worse?
The self check is at average:0.31, maximum:8.12 RMS 0.51 while with large lut test chart it’s at: 0.03, maximum 0.86 RMS 0.11
Left is premade 49xx test chart while right is 49xx auto optimized in picture
The self-check doesn’t tell you which of the profiles is more accurate, it just tells you which one matches its test set better as a means of diagnosis (anything below avg of 0.5 and peak 5 is fine generally speaking). For testcharts with even distribution of points (“regular grid”), it is expected that the self-check dE is lower than for a perceptually distributed set, but it is not a measure of profile accuracy.
The self-check doesn’t tell you which of the profiles is more accurate, it just tells you which one matches its test set better as a means of diagnosis (anything below avg of 0.5 and peak 5 is fine generally speaking). For testcharts with even distribution of points (“regular grid”), it is expected that the self-check dE is lower than for a perceptually distributed set, but it is not a measure of profile accuracy.
If I want to verify my 3DLUT that I made with MadTPG do I just select xxxl video verification chart without using simulation profile? When I did that it reported a bunch of colours with a dE of 2.5-3.5, and a average of 0.5dE is it possible to increase the accuracy of colours to around 0.3dE, and maybe improve the ones that where around 3dE to under 1 dE?
Or was I doing the verification wrong, and it didn’t use the 3DLUT?
Update: hmm running the verification with rec709 simulation profile and 3DLUT box ticked, with black output offset 100% 2.4 clearly gives me wrong values as it seems it’s applying the tone curve calibration twice.
Okay seems im not supposed to use: Use simulation profile as display profile, it mostly gave me correct gamma tracking this way but clearly it went wrong somewhere as grayscale value 55% shows 2.6 gamma, do i need to redo the profiling or the actual calibration as it pretty much follows the calibrated curve if i unselect simulation profile, and verify against icc profile only.
Hmm there’s still something weird with the some of the yellow colors, i did a second calibration, and profiling i still get high dE in blues, but the yellow colors that report a de of around 0.2-0.4dE clearly looks distinguishable even if the reported dE says it shouldn’t
Here’s a HTML report of the verification the yellow patches from around 225ish clearly looks wrong
