Originally Posted by EvLee
Generally speaking, it is advisable not to preserve all the saturation when performing tone mapping. Tone mapping is about making tradeoffs in multiple dimensions of color appearance, and sacrificing some saturation allows you to better maintain the local relationship between colors as well as image detail in highlights. I have tested this quite a bit over the years with multiple colorists, DPs, etc... and after being presented the options we always settle on a tone mapping that sacrifices some saturation for lightness. The direction (increase vs decrease) of saturation correction depends on what tone mapping operation you are using. This has been supported by color science research (see: https://www.cl.cam.ac.uk/~rkm38/pdfs/mantiuk09cctm.pdf
, and http://www.erikreinhard.com/papers/cic_2013.pdf
I've tried the algorithm suggested in the 2nd paper - but using it doesn't produce the slighest visible change for me!
I'm wondering if this is due to either:
a) Maybe it's a problem that I use ICtCp instead of IPT? But in theory ICtCp should be superior to IPT.
b) Maybe my tone mapping is already handling saturation in the way the algorithm aims for? FWIW, I'm doing tone mapping in ICtCp, and I am already doing a desaturation according to the way Dolby and SMPTE 2390 suggest. So maybe there's simply nothing left to do for the algorithm? I'm totally not sure about this, though.
Originally Posted by EvLee
Hue preserving tone mapping is pretty much always desirable. There are various ways to go about it. The thing to watch out for is what space you use to calculate hue, and how the color volume in that space transforms back into display RGB. Performing hue correction in a perceptual space like ICtCp and then converting back to RGB can cause colors to rotate into clipping, so you want to check in your display RGB space that nothing is getting clipped.
Yes, in highest quality mode I'm doing everything in ICtCp, and it's a bitch to work with. I think I have the clipping problems solved, but it's costing quite a bit of GPU performance.
Originally Posted by markmon1
You can output this to a raw file? Like a yuv file or such? I can see if any of the ffmpeg builds I have support it. It looks like maybe pix_fmt yuv444p16le may read it.
If this is helpful you can try ffmpeg: https://www.convertthisfile.com/files/ffmpeg.exe
Something like ffmpeg -f rawvideo -pix_fmt yuv444p16le -s:v 3840x2160 -r 25 -i inputfile.yuv -c:v libx265 outputfile.mkv
where s:v is resolution and -r 25 is frame rate
ffmpeg -pix_fmts will list all the formats.
Thanks, I'll give that a try, when I find some time.
Originally Posted by Fer15
I’ve been following this thread out of interest and would just like to take a moment to express my appreciation for all the work that you guys are doing.
I’m not sure if you’ll find the following useful to the current discussion or not, but I would just like to provide a few more examples of the differences between the scientific and dumb mode when playing back further examples of real content.
I noticed that there were some UHD titles, which appeared to be yielding some questionable tone mapping effects when the scientific approach was enabled. Some of the Nolan UHD collection struggled in particular (The Prestige, Batman Begins, Inception). As you can see from the screens, Caucasian skin tones took on a redder/ more golden effect when the scientific approach was used. Furthermore, whites - such as the clouds in the sky, or snow also became a warm white to the point that they appeared almost a light yellow.
Thanks for the comparison screenshots! Could you please provide timecodes for each of these screenshots? Without timecodes it would be *really* painful for me to analyze all these frames, because I'd have no easy way to double check this myself. I suppose these are all USA releases? Usually they have the same color grading as releases from other countries, but sometimes the color grading might differ, so it'd be good to know.
Originally Posted by KarlKlammer
I think Javs has a talent to find scenes that put every tone mapping to shame.
OK, here are screenshots of the spear and two from Mad Max. I took two pictures from each scene. One is with the Radiance Pro Intensity Mapping (IM) and one with a LightSpace ST.2084-LUT (LS).
IM makes it almost impossible to capture the spear right. But the difference between yellow and green spots is really too subtle to catch.
The first scene from Mad Max doesn't look that horrible with IM in reality. But the odd color reproduction of the lightning and flame is clearly visible.
And the pink smoke in the second Mad Max picture might be a little bit off too.
Thank you for the screenshots! Ok, so we can conclude that:
1) Lumagen's IM does not desaturate at all. It's the same approach which madVR uses if you select the option "100% luminance reduction, 0% saturation reduction". IMHO that's not a good choice, to be honest. As you can see, it produces highlight detail loss in some scenes. Ok, these scenes are extreme cases, but they do occur.
2) LightSpace seems to use "dumb/naive" color handling, which is surprising to me.