DeepMerge will actually just Merge all of the deep samples, I don't believe it resamples the data unless you tell it to drop fully hidden samples ( alpha >= 1 ). I'm not %100 how The Foundry have implemented this but my guess is that the samples are appended to the end of a list vs. interleaved and you only get the sorted samples when requested via the Deep API.
DeepToImage doesn't average, it composites all of the samples together flattening the deep data. Think of each pixel like a mini comp and you're merging the samples front to back, if the first sample has an alpha of 1.0 then you're not going to see any of the others. Cheers On 2012-12-11, at 2:35 PM, MathieuLeclaire <[email protected]> wrote: > Hi everyone, > > I'm a programmer and I'm new to Nuke. I'm trying to implement a deep image > compositing pipeline at my studio and I'm trying to make Nuke 7 deep image > compositing tools for Open Exr 2.0 work, but it's not giving me the expected > results. > > Nuke reads our Open Exr 2.0 files. If I use a DeepSample node, I can see that > all the information is there. I have a list of distances and rgb colors per > sample for each pixel. > > If I try to view the data in the viewer, it gives me a very pixelated result > as if there was no anti-aliasing. All my tests points to Nuke always only > showing the closest of the samples instead of an average of the list of > samples. The problem becomes more evident when I use a DeepMerge. > > I was under the impression that the functionality of a DeepMerge followed by > a DeepToImage was to : > > A) DeepMerge > 1 - Per pixel, resize the list of deep samples from each input images to the > smallest common denominator (lets call it n). > 2 - Merge all the lists together (always on a per pixel basis.) > 3 - Sort the list by depth. > 4 - Keep only the first n values of that merged and sorted list. > > B) DeepToImage > 5 - Process the average color of that new list. > > > But from all the tests I have made, it always seems to return the front most > sample color instead of an average of colors. I've also tried using the > DeepRecolor to plug the aliased result (one color per pixel instead of a > color per sample) but I still get the same phenomenon. > > Take a look at the attached image to see what I mean. > > I've also tried playing with the very limited options in the various deep > nodes like "volumetric composition" and "drop hidden samples" etc. but > nothing changed the results. I've tried forcing the images to all have the > same number of samples per pixel, thinking maybe Nuke assumes all pixels from > each images all have the same number of samples, but it didn't change nothing > to the results. > > What am I doing wrong? I'm at a lost. Any input would be greatly appreciated. > > Thanks. > > > > Mathieu Leclaire > Director of R&D > Hybride Technologies, a Ubisoft division > <DeepMergeBugs.jpg>_______________________________________________ > Nuke-users mailing list > [email protected], http://forums.thefoundry.co.uk/ > http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-users _______________________________________________ Nuke-users mailing list [email protected], http://forums.thefoundry.co.uk/ http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-users
