search

Beep6581 / RawTherapee

A powerful cross-platform raw photo processing program
https://rawtherapee.com
GNU General Public License v3.0
2.81k stars 318 forks source link

Custom dcp profile for Nikon-D750 #2682

Closed Beep6581 closed 9 years ago

Beep6581 commented 9 years ago

Originally reported on Google Code with ID 2700

 .. better late than never :(

DrSlony provided a set of color checker passport raw samples and custom dcp profiles
built on these samples. 
Two of them gave consistently good measures over the various samples.

Although there is a strangely large error on grey patches of the provided samples so
I crosschecked on Dpreview's ISO100 shot where the measures are much better with no
problem at the grey scale patches. 
It could be a problem with the specific passport and :( and I think for some samples
a bad role played by the overexposure with which they were shot.
The cloudy_wl16300 is a rebuilt after hacking the DNG's WL from 15500 to equal with
RT's 16300 (this 16300 comes from inspecting raw histograms) to overcome some of the
possible problems with overexposure and indeed it gave a bit better measures than cloudy.

Attached are
- the winner sunny.dcp de2000 avg 1.21 max 2.96 (light skin) measured on dpreview's
sample
- and the close second cloudy_WL16300 avg 1.26, max 3.18 (light skin) measured on dpreview's
sample

a list with the probability of detecting de2000 errors is 
https://code.google.com/p/rawtherapee/issues/detail?id=2368#c11

Only the light skin looks problematic and this looks like a systematic error with profiles
built by x-rite colorchecker passport utility as used by RT. I see it on most dcps
:).

Many detailed per color results as png screenshots 
https://drive.google.com/folderview?id=0B0NqktTgc54sYzRCWU10LWZvSFU&usp=sharing
Look only at the last column ..

The settings used are neutral 
+ hot/dead pixel filters (to remove outliers)
+ noicm output (max precision on 8bit data, RT_sRGB looks to be a bit less precise).

+ WB custom, picked on the mid grey (neutral5) patch as according to x-rite's reference
data, this is the most neutral grey of the patches

Reported by iliasgiarimis on 2015-03-01 00:07:14

Beep6581 commented 9 years ago 
Are you OK with having Nikon_D750_sunny.dcp committed?

I'm intrigued by what you wrote,
"noicm output (max precision on 8bit data, RT_sRGB looks to be a bit less precise)"
RT_sRGB is supposed to be more precise than vanilla sRGB with four times as many points
in the LUT.

Reported by entertheyoni on 2015-03-04 20:19:18

Beep6581 commented 9 years ago 
Yes I vote for commiting Nikon_D750_sunny.dcp . It gives a very good score consistently.

This "noicm" setting is actually Olli's (Oduis who is the author of colorchartcheck)
recomentation. As we use this utility to measure the exported patches I prefer to use
what is recommended for this 
http://rawtherapee.com/forum/viewtopic.php?f=1&t=2950&start=570#p34966

As I understand it, every LUT should be less precise than the function it replaces.

Then comes LCMS which does (or not) a perfect job also. It's a matter of cooperation
between RT and LCMS.

Jacques presented the result of RT_sRGB.icc on 16bit file and there he showed it to
be better. But what happens on 8bit files ?. How is LCMS dealing with a 10bit LUT on
8bit data ?. 

We can explore it by using a reference target 
http://cl.ly/3F2v0T0j2x3P/RGB_255to0_52steps_Bayer.dng.zip
http://www.dpreview.com/forums/post/50706883
http://www.dpreview.com/forums/post/50707919 .. I'll convert the 16bit linear values
to sRGB gamma and check later ..

Reported by iliasgiarimis on 2015-03-04 21:41:48

Beep6581 commented 9 years ago 
Committed, thank you.

Reported by entertheyoni on 2015-03-11 14:26:06