Re: [Openexr-devel] Color Management Proposal / Siggraph meeting

[Florian Kainz]

The inputMedium attribute is not necessary for image processing.
Its only purpose is to store where the image came from (for example,
to help a person understand limitations of the input medium, as
you suggested).  Computer-generated images that are generated
directly in scene-referred form, or composites images assembled
from several elements that were imported from different input media,
would not typically have an inputMedium attribute.

During my presentation at Siggraph, Charles Poynton and Greg Ward
pointed out that the sceneReferredSpace attribute might be redundant;
the existing chromaticites attribute should be enough (unless the
scene-referred space has more than three channels).

The outputMedium and referenceDisplay attributes describe what the
image should look like when it is shown to a paying audience.  The
raw scene-referred pixel data in the OpenEXR file specify what your
(real or virtual) scene looks like, not the amount of light that
comes from the theater screen.
You can think it this way: By processing and compositing scene-
referred images you assemble a virtual scene.  By adding outputMedium
and referenceDisplay attributes, you give instructions to image
viewing and recording software, something like "make the image look
as if my virtual scene had been photographed on film stock X, and
printed and projected using process Y."  Changing the outputMedium
or the referenceDisplay attribute changes the look of the image.

Florian


Thad Beier wrote:
> Florian Kainz wrote:
>
> > ILM's OpenEXR color management proposal, presented at the
> > Siggraph 2004 "OpenEXR, Film and Color" Birds of a Feather
> > meeting, is now available on the OpenEXR web site:
> >
> >     http://www.openexr.com/OpenEXRColorManagement.pdf
> The document makes perfect sense.  I agree completely with ILM's
> position that OpenEXR files should be scene-referred (although I
> would certainly support a less awkward name.)  All other colorspaces
> can be generated from the scene-referred data, but none of them
> deserve preference over the real world information.
>
> That said, I don't understand why the inputMedium, outputMedium,
> or referenceDisplay attributes are part of the OpenEXR file.  What
> use could they possibly be put to?  How are they not a anachronistic
> throwback to previous device-dependent image file formats?
>
> When an image is created from an input medium of some kind, it is
> converted into a scene-referred OpenEXR file.  The only possible
> use I can see of remembering the inputMedium name is to help understand
> what the gamut and dynamic range limitations of the inputMedium was -- but
> two images of the same scene input from two different media should be
> identical (within the accuracy that the IMSRs represent reality.)  I
> wouldn't think that there would treat those two files any differently in
> any further processing.
>
> That said, everything else makes perfect sense.  The devil is in the
> specification of these transformations, especially as they apply to
> film stocks.
>
> Thad Beier
> Hammerhead Productions
> address@hidden
> 818-762-8643
>
>
> _______________________________________________
> Openexr-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/openexr-devel