Great to have someone from Arri here! Unfortunately that post left me more confused than I was before :)
To start with, when I said "Arri's own Log C LUT (as well as the original Cineon LUT), are really designed for video output", I wasn't referring to the LogC to REC 709 LUTs. I meant that the concept of a black point is related more to output-referred rather than scene-referred image states. When using the LogC or Cineon formulas, anything below 95/1023 will be mapped to negative values. No scene-referred encoding of a physical scene should ever contain negative values. No matter how much one increases the exposure of such an image, the negative data can never be regained. I'm not saying that the LogC Log to Lin is wrong in any way, I'm just saying it's more geared towards making a pretty picture (when viewed through the appropriate monitor LUT) than it is to giving color values suitable for VFX work. I therefore expressed an interest in using the Josh Pines math for reading these images instead. The question was what parameter values would best decode the LogC data. As is stated in that pdf you referred to, 18% gray is mapped to 400/1023. The other things that one needs to know is the negative gamma and negative density per log code value. These are of course film properties (as the formula is designed for cineon film scans), but unlike pseudo-log formats (such as FilmStream, ARRI Log F, Panalog, S-Log) Log C should have corresponding values. Assuming a density of 0.02, some empirical testing gives a negative gamma of 0.45 to match the look of an image converted by the ARRI equation. This is what we are using on our current production. We can leave that question for now though, as reading the documents you recommended has left me with new questions :) You say that the math in the ALEXA Log C Curve pdf is correct, and the one used in Nuke. You also say that the math in the Alexa Color Pipeline for Nuke pdf is not up to date, and should not be used. But as far as I can tell, it's this latter formula that is actually used in Nuke (6.3v2)! Another interesting discrepancy is the ALEXA Wide Gamut RGB primaries. When using the Nuke Colorspace node to convert from the Alexa primaries to CIE (or anything else), one gets a somewhat different matrix than is given in the ALEXA Log C Curve pdf. If The Foundry has gotten this wrong, perhaps someone should inform them? Andreas Bravin Karlsson Compositing Supervisor
_______________________________________________ Nuke-users mailing list [email protected], http://forums.thefoundry.co.uk/ http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-users
