nice thanks for doing these! this seems to confirm the white balance hypothesis… dcamprof normalises wb to D02 (apparent by the 0 DE, right?). with that out of the way it’s ballpark the same… a bit worse which is expected, but not by much (3.29 vs 2.76 on the worst patch).
so i guess i’m looking for the right illuminant here…
For the above runs, I used the dcamprof default D50 illuminant. I just ran StdA, and it dropped the max dE to 3.40, not a significant difference.
well okay you shine a different light on the patch, go through the observer and then you’ll get slightly different metameric behaviour here. but to compare to the reference it’ll first thing white balance it to make the D02 80% white patch match the reference values. this transform doesn’t count as error.
in my case i’d like to characterise this absolutely, without the wb step, so i can create a mapping camera → cie observer that only depends on the cfa curves (one mapping for each given illuminant). and using the monochromator curves vs. the predicted here gives me a colour cast (that i assume dcamprof normalises out in the wb step).
(but now i’m all confused again with all the different steps and errors. could it be the issue that they use the abbridged range 400-700nm only?)
I wouldn’t think so, the camera responses through the CFA go to zero at about these limits.
I haven’t studied Anders’ code so couldn’t tell you. Further, I’m not that versed in the math behind all this, alls I know is he generates an illuminant-specific LUT from the same SSF set.