My interest is in using QC for still-image (not motion-based)
compositions that end up being rendered into a specific bitmap size.
(Actually, I know there are more of us who are interested in using QC
for this.) It has taken me quite a while to get used to the "pull"
nature of things and the fact that for the most part, you need to
think in terms of units rather than pixels. It still boggles my mind
a bit.
Yes, Quartz Composer, being OpenGL based, works in a 3D referential
which has to be resolution independent, therefor the use of "units".
If you are doing image manipulation through Core Image, then
coordinates are in pixels since the filters work on the pixels of the
image and not on the final rendering destination (i.e. the 3D
referential).
Quartz Composer provides several patches to convert between pixels and
units when it's necessary like Image Dimensions, Rendering Destination
Dimensions...
