Re: [Gimp-developer] Histograms in unbounded mode sRGB

[Gez <listas ohweb com ar>]

El 2014-04-22 15:25, yahvuu escribió:
> Hi,
>
> Am 18.04.2014 22:10, schrieb Elle Stone:
> > I think the only reasonable solution is to keep the WideGamutRGB
> > chromaticities available for use as an editing/compositing color 
> > space,
> > and use that color space to display histogram information (and also to
> > display Color Picker and Color Selector information).
>
> please allow me to make the case for using the color space of the 
> respective
> export file format for histogram displays.
>
> You recently gave striking examples of how working with RGB _numbers_
> can quickly
> become unwieldy from a user point of view. This probably applies as 
> soon as the
> limitations of the traditional 8-bit sRGB processing are relaxed. There 
> is
> nothing wrong with RGB color models, it is just that the numbers can 
> become
> difficult to interpret for human beings.
>
> So, as a thought experiment, i'd like to get rid of any kind of RGB
> triples in the
> UI. To see where it hurts the most. This will be the places where RGB
> numbers are
> really needed. After all, GIMP is about colors, not numbers.
>
> Say, we get an adobeRGB camera image and the task is some touch-up and 
> warping.
> The deliverables are 1) a JPG for the web and 2) a proPhotoRGB file
> for that mega
> stock company.
>
> I find that most of the editing can be performed without looking at a 
> single RGB
> triple. Image transforms do not expose RGB numbers, neither do most of
> the filters.
> Even many of the tools in the Colors menu do not refer to RGB numbers.
> Of course,
> this is different for levels/curves. But by large, these tools are
> used on the 'value'
> channel, not on the distinct R,G,B channels. Here, working on the
> luminance channel
> instead would probably be an improvement.
>
> I find it is only on *export* when the RGB numbers become really
> important. Much like
> output-specific scaling and sharpening, each of the deliverables needs 
> specific
> color treatment.
>
> Here, the histograms need to show the R,G,B channels of the specific
> output color
> space to allow assessment and correction of the conversion. Probably,
> this is also
> where the curves/levels tools should be working in the output color 
> space. For
> example, how else could someone boost the midtones without adding
> clipping -- when
> maximum and minimum range of the curve do not refer to the actual range 
> of the
> output file format? (not even talking of non-matching color primaries 
> here)
>
> These color modifications that are specific to the output files are
> hot candidates for
> being stored in export pipelines, again analogous to scaling and
> sharpening operations.
>
> I'm less sure in how far this is an analogy to CMYK export -- is an
> equivalent to
> the 'press projection' needed here? Or is it sufficient to set the RGB
> color space
> of histogram/curves etc. to the currently active softproofing color 
> space?


This is no doubt an interesting thought experiment, but I'm afraid it 
can't live much beyond it.
The way users interact with GIMP is too complex to do something like 
that.
Your example cherry-picks situations where the color part can be left 
for the final stage, but there are several manipulations where you start 
by doing some color correction.
And since RGB is how digital color images are stored, having tools for 
watching what's going on in channels while you edit (like histograms, 
waveforms, etc.) is essential for manipulating color with accuracy.
Destroying image data inadvertently is easier than you think, specially 
when you're manipulating data that doesn't fit in your output device's 
gamut.

And all this things start to sound like squaring the circle. Again, it's 
an interesting thought experiment, but if we need thought experiments to 
make a model fit, breaking the existing paradigms of image manipulation, 
then there's probably something wrong with the model.

I'm not against different ways of doing things, but it seems like making 
the unbounded RGB model fit requires to re-think a lot of the existing 
structure, not only the internals of GIMP but also how users use the 
tool.