Hi Ivan, If you think of each deep sample as a light absorber of a unit length then each sample has an absorbance amount which can be derived from the alpha. So 1 - alpha = transmission. Absorbance = -ln( transmission ). ( from beer lambert ).
DeepRecolor rebalances the alpha values by considering the relative absorbance of each sample rather than the relative alpha. So: orig alpha: 0.52 targe alphat: 0.9 orig absorbance: -ln( 1.0 - 0.52 ) = ~0.733 target absorbance: -ln( 1.0 - 0.9 ) = ~2.302 gain = target absorbance/orig absorbance = ~3.137 Samp1 : alpha = 0.4 Samp2 : alpha = 0.2 Samp1: absorbance = -ln( 1.0 - 0.4 ) = ~0.510 Samp2: absorbance = -ln( 1.0 - 0.2 ) = ~0.223 Samp1: target absorbance = ~ -0.510 * 3.137 = ~ -1.602 Samp2: target absorbance = ~ -0.223 * 3.137 = ~ -0.700 Samp1: target alpha = 1.0 - exp(-1.602) = ~0.798 Samp2: target alpha = 1.0 - exp(-0.700) = ~0.503 The last few calculations can be simplified as well. If we consider, target transmission = exp( -gain * absorbance ) and, absorbance = -ln( transmission ) then new target transmission is: exp( gain * ln( transmission ) ) which is just, pow( transmission, gain ) Also for reference, there's a couple of great papers on the exr site by Florian and Peter Hillman which go into great detail on what deep samples represent and how to manipulate them. http://www.openexr.com/documentation.html ( links are at the bottom of the page ) Cheers, Jon. On Mon, Dec 16, 2013 at 8:33 PM, Ivan Busquets <ivanbusqu...@gmail.com> wrote: > Thanks Steve, > > In this case, I'm talking opacity only. > > Color values seem to follow the same logic as non-retargeted recoloring. > That is, divide by the input alpha, and multiply by each sample's alpha. > > It's the re-distributed alpha values that I can't find the secret sauce for > :) > > > > > On Mon, Dec 16, 2013 at 12:29 PM, Steve Booth <st...@boothsoft.com> wrote: >> >> Ivan, >> >> >> >> The luminance equation comes to mind: >> >> >> >> Y = 0.2126 R + 0.7152 G + 0.0722 B >> >> >> >> I'm wondering if the Recolor is trying to maintain essentially the same >> luminance during the recolor operation. Thus, for example, if you are >> shifting to the blue, you might actually want a higher opacity? >> >> >> >> Just a guess... >> >> >> >> Steve >> >> >> >> >> >> From: nuke-dev-boun...@support.thefoundry.co.uk >> [mailto:nuke-dev-boun...@support.thefoundry.co.uk] On Behalf Of Ivan >> Busquets >> Sent: Monday, December 16, 2013 12:18 PM >> To: Nuke plug-in development discussion >> Subject: [Nuke-dev] How is DeepRecolor distributing a target opacity >> amongst samples? >> >> >> >> Hi, >> >> Not strictly a development question, but I figured there would be a better >> chance of someone in this list knowing what's happening under the hood here. >> >> Long story short, I'm trying to implement a procedure to re-target opacity >> of each sample in a deep pixel, akin to what's happening in a DeepRecolor >> node when "target input alpha" is checked. >> >> I've got this to a point where it's working ok, but I think I might be >> missing something, as my results differ from those you'd get in a >> DeepRecolor. >> >> My re-targetting algorithm is based on the assumption that the relative >> opacity between samples should be preserved, but DeepRecolor clearly uses a >> different approach. >> >> Example: >> >> Say you have a deep pixel with 2 samples, and the following opacities: >> >> Samp1 : 0.4 >> >> Samp2 : 0.2 >> >> The accumulated opacity is 0.52 (Samp1 over Samp2). Note that Samp1 >> deliberately has an opacity of 2 times Samp2. >> >> Now, let's say we want to re-target those samples to an accumulated >> opacity of 0.9. >> >> What I am doing is trying to calculate new opacities for Samp1 and Samp2 >> in such a way that Samp1 == 2*Samp2 and Samp1 over Samp2 == 0.9 >> >> This gives me the following re-targeted values: >> >> Samp1 : 0.829284 >> >> Samp2 : 0.414642 >> >> >> >> I'm happy with those, but it bugs me that DeepRecolor throws different >> results: >> >> >> >> Samp1 : 0.798617 >> Samp2 : 0.503434 >> >> Which meets the Samp1 over Samp2 == 0.9 criteria, but does not preserve >> the relative opacities of the original samples. >> >> It seems to me like DeepRecolor is applying some sort of non-linear >> function to apply a different weight to each of the original samples, but I >> haven't been able to figure out the logic of that weighting, or a reason why >> it's done that way. >> >> Does anyone have any insight/ideas on what DeepRecolor might be doing >> internally? >> >> Or a reason why you might want to distribute the target alpha in a >> non-linear way? >> >> >> >> Thanks, >> >> Ivan >> >> >> >> >> >> >> _______________________________________________ >> Nuke-dev mailing list >> Nuke-dev@support.thefoundry.co.uk, http://forums.thefoundry.co.uk/ >> http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev >> > > > _______________________________________________ > Nuke-dev mailing list > Nuke-dev@support.thefoundry.co.uk, http://forums.thefoundry.co.uk/ > http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev > -- Jon Wadelton, Nuke Product Manager. The Foundry 5 Golden Square, London, W1F 9HT Tel: +44 20 7479 4350 • Web: www.thefoundry.co.uk The Foundry Visionmongers Ltd • Registered in England and Wales No: 4642027 _______________________________________________ Nuke-dev mailing list Nuke-dev@support.thefoundry.co.uk, http://forums.thefoundry.co.uk/ http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
