Fred Miranda Offline Admin Upload & Sell: On
|
Raamiel wrote:
About the Adobe "Camera" profiles on ACR and lightroom, I would raise the explanation on architectural structure differences between Adobe and Cobalt.
That's the first time for me here, I am the creator of the Cobalt project and the tech engineer of the project.
To better understand how profile work in the DNG development pipeline, it's mandatory to start with logic fundamentals: "Camera" profiles are DCP profiles and as any other DCP can be split into different stages.
this is a typical sample of profile "Camera" anatomy:
{
"UniqueCameraModel": "Canon EOS 5DS R",
"ProfileName": "Camera Neutral",
"ProfileCopyright": "Copyright 2014 Adobe Systems, Inc.",
"ProfileEmbedPolicy": "Embed if used",
"ProfileCalibrationSignature": "com.adobe",
"CalibrationIlluminant1": "StdA",
"CalibrationIlluminant2": "D65",
"ColorMatrix1": [
[ 0.684800, -0.166100, -0.022100 ],
[ -0.390400, 1.093100, 0.343400 ],
[ -0.047000, 0.125100, 0.603900 ]
],
"ColorMatrix2": [
[ 0.625000, -0.071100, -0.080800 ],
[ -0.515300, 1.279400, 0.263600 ],
[ -0.124900, 0.219800, 0.561000 ]
],
"ForwardMatrix1": [
[ 0.797800, 0.135200, 0.031300 ],
[ 0.288000, 0.711900, 0.000100 ],
[ 0.000000, 0.000000, 0.825100 ]
],
"ForwardMatrix2": [
[ 0.797800, 0.135200, 0.031300 ],
[ 0.288000, 0.711900, 0.000100 ],
[ 0.000000, 0.000000, 0.825100 ]
],
"DefaultBlackRender": "None",
"BaselineExposureOffset": -0.170000,
"ProfileLookTableDims": [ 90, 16, 16 ],
"ProfileLookTableEncoding": "sRGB",
"ProfileLookTable": [
{ "HueDiv": 0, "SatDiv": 0, "ValDiv": 0, "HueShift": 0.000000, "SatScale": 1.000000, "ValScale": 1.000000 },
{ "HueDiv": 0, "SatDiv": 0, "ValDiv": 1, "HueShift": 2.050087, "SatScale".........
In the first stage, there is the model ref and luma type.
Colormatrix give the approximation of WB , Hue and color temp. to the software.
Fowardmatrix are the colorimetry reference and are used as information and data by the raw decoding process.
That is followed by the Looktable , working for the color grading and the latest is the contrast curve.
If we inspect the data we find the first issue; Forward Matrix are weird, are the same so the profile IS NOT dual illuminant.
And the multiplier is connected to the transformation of ProPhoto, specifically from RGB to XYZ.
So the characterization is missing and this job is done by the LookTable where characterization and color correction get melted.
The sad thing is the characterization is before every user control (shadow recover, exp) and Looktable is after them.
So on this structure, there are two problems, first is the profile is not a dual illuminant profile.
The second even worse, characterization process is done at the wrong point of the processing pipeline.
The result is the "camera" profiles, no matter what camera profiles, are "broken" just after any user control changes (again recovering highlights, opening shadows..exp..etc) .
Cobalt emulations are based on DCP base so dual illuminant.
The colorimetry stage consists of specific ForwardMatrix which incorporate an improved CAT compared to the Bradford; followed by HueSatMap charts specific for any different illuminant.
So on this structure the characterization stage and the color grading, are well separated and on the right part of the development pipeline.
So the funny thing is the Cobalt digital emulation are better than the original just because based on correct profiles....Show more →
I'm glad the Cobalt team does not work for Adobe, otherwise we would not be having this discussion. 
|