Home › Forums › General Discussion › Comparison of 3DLUTS made with increased Pattern Count
- This topic has 5 replies, 3 voices, and was last updated 7 years ago by Kapkan.
2016-03-03 at 9:00 #2086
Is there maybe a comparison to see how much better the 3DLUT becomes when increasing the amount of patterns?
I´m still not sure if its worth a try.
Basti2016-03-03 at 11:09 #2092
This depends on the specific situation (i.e. the display, playback setup etc). Typically diminishing returns after roughly 1500-2000 patches, and personally I’m seldomly going over 1500 (most of the time I’m using around 1000). It’s unlikely you’re going to perceive a visual difference, although the delta E when doing verification may go down.2016-03-03 at 15:13 #2099
On my TV (Samsung CCFL from 2009), in PC 4:4:4 mode where colorspace cannot be adjusted through TV controls, 100% primary colors saturation sweeps were off by dE 1.5, 2.3, and Blue was off by dE 3.3 after Calibration + Profiling using 2681 Auto-optimized patches, even though the original (un-calibrated) colorspace wasn’t that far off. 25%, 50%, and 75% sweeps were all below dE 1. Skin colors were also accurate. I noticed the same post-calibration results pattern after creating a 3DLUT for my monitor (Eizo Foris FG2421). Again 2681 patches was not enough to get 100% saturation sweeps accurate (within dE 2). Some year ago I recall having to make custom patches set to get those 100% sweeps to have accurate results when using 2681 patch set. To get those 100% saturation sweeps accurate on my monitor, I had to use 4289 Auto-Optimized patches, but original colorspace (un-calibrated) was awful, where 100% red saturation sweep results showed dE of 8.7. Even after 4289 Auto-Optimized patches, one of skin tone dE’s was 2.3. I don’t have the actual HCFR measurements numbers (haven’t kept them), but I do have the pictures, which aren’t 100% accurate like actual dE numbers, but I do recall those numbers being around dE 2.3, always higher for 100% saturation sweeps. Take a look at 100% green and red saturation before and after. 3DLUT slightly cut off green and red coverage below of where it could. Again, HCFR pictures don’t depict everything accurately…
I do realize that 100% saturated colors are very rarely used in films, but they are more often used in games. Maybe Auto-Optimized setting could benefit from some custom adjustments to make 100% saturated colors accurate?
2016-03-03 at 15:23 #2102
- This reply was modified 7 years, 1 month ago by Kapkan.
Errors in the 100% saturations are more likely than not coming from gamut limitations of the display in question. If the display can’t fit the whole of Rec. 709, this means colors near the gamut boundaries will have to be clipped. There’s not much that can be done about these display hardware limitations, and increasing the patch count won’t help.2016-03-03 at 15:24 #2103
The image wouldn’t show up when inserted while typing in Text tab. Here’s the link – https://dl.dropboxusercontent.com/u/93831976/Other/FG2421-1Dvs3D.jpg .
2016-03-08 at 0:39 #2249
- This reply was modified 7 years, 1 month ago by Kapkan.
Errors in the 100% saturations are more likely than not coming from gamut limitations of the display in question. If the display can’t fit the whole of Rec. 709, this means colors near the gamut boundaries will have to be clipped. There’s not much that can be done about these display hardware limitations, and increasing the patch count won’t help.
Yes, true, and they are often considered the least important out of all possible color saturation levels. However, I most certainly remember correspondence between you and I on AVS Display Forums about a year (if not more) ago regarding creation of a custom Test Chart where 100% primary and secondary color saturation measurement patches were either specifically selected or weighted heavier. Calibration performed using that custom Test Chart, which I no longer have, resulted in 100% primary and secondary color saturation sweeps having dE ~1.2 (dE 2000) or less. With DisplayCAL’s new Auto-Optimized Test Charts I wasn’t able to achieve such low dE’s even with 2681 patches on my TV, which does cover 99%+ of sRGB/Rec.709 and absolutely can achieve much better 100% saturation sweep accuracy with a custom chart that specifically dictates measurement and correction of 100% primary and secondary saturation sweeps or at least places more weigh into those area.
The image I provided in my earlier reply was for my monitor, not TV. Unlike my TV’s 99%+ sRGB/Rec.709 coverage, my monitor’s sRGB / Rec.709 coverage was only about 93% after 4289 patch 3DLUT calibration. Of course, I was not expecting low dE’s for 100% primary and secondary color saturation sweeps with such poor colorspace coverage, even though I did achieve incredible improvement in those 100% primary and secondary color saturation sweeps, but only after using 4289 patches. For example, my monitor’s native 100% red saturation sweep was about ~dE 8 with 1DLUT alone. After performing a calibration with 2681 patches, the resulting 3DLUT showed that 100% red saturation sweep dE fell down to about dE ~4.0 (huge improvement), but after 4289 patches, it fell further down to about dE ~2.5, a less drastic change, but still a considerable and visible improvement.
Even with dE 2.5 (dE 2000), it is quite easy to spot that my monitor’s 100% red saturation has some pink in it. With that I am implying that the whole “below dE 3 is very hard to notice” is not true for everyone, especially since there are different methods of measuring dE. The most recent and the most-frequently used method, dE 2000, is the most relaxed one too. To my eyes, dE 2000 error of dE 1.5 or less is where the difference become negligible. I could tolerate dE 2.00 at best using dE 2000 formula, but above dE 2.0 I consider “inaccurate”. That’s for colorspace, but for grayscale my eyes are even more sensitive.
Is there any way you could, PLEASE, create or tell us how to create a custom Test Chart using already existing Auto-Optimized settings, BUT with forced measurements for 100% color saturation sweeps? It would add at least 6 new patches + whichever other necessary patches to accurately adjust 90-100% color saturation sweeps.