Am 30.11.2010 23:58, schrieb Christian Froeschlin:
>> Now what do I get? Not a straight, smooth, gradual gradient from
>> light to dark but a parabolic curve of values from light to slightly
>> darker but still light to abruptly skewed to black.
>
> I have to admit I don't understand this either. Until now I
> thought that since 3.7 now uses a linear color space, 0.5 would
> represent a true midgray and all the nifty gamma handling ensures
> that everyone perceives that midgray on their display when viewing
> the image file. So I'd also have expected the end result of your
> test scene to visually yield evenly spaced brightness steps.
>
> Looking at the png itself it seems the upper two rows use 127
> for midgray while the lower uses 186. From all I heard so far
> 186 should appear as midgray on a calibrated gamma 2.2 display
> so that would appear to be correct. But I too perceive the
> upper two rows as more visually correct. It's as if my lcd
> used linear color space. I wonder if the Windows 7 built-in
> display calibration is somehow playing tricks on me.

No, it's your eye playing tricks on you.

Our eyes are highly calibratible optical measurement tools - they work 
as well at a brightness of 10% as at 1,000% (and then some), and get 
object colors right regardless of lighting conditions.

There's the key.

Grab a piece of 100% white cardboard and another piece of 90% gray 
cardboard.

Now dim the light to half brightness. Obviously, your white cardboard 
will now reflect only 50% of full-brightness white into your eye, and 
your gray cardboard will reflect only 45% of full-brightness white.

So the absolute difference between the two pieces of cardboard has 
diminished from 10% full brightness to 5% full brightness. But our eyes 
are designed and trained to look not for the absolute light intensity, 
but for /relative/ difference, so that we can identify the "pigment" of 
objects irrespective of lighting conditions. The /apparent/ brightness 
difference between 45% and 50% is therefore roughly the same as between 
90% and 100%.

As a result, a truly linear gradient doesn't /appear/ linear to us: The 
"distance" between 10% and 20%, for instance, is percieved as roughly 
equal to that between 50% and 100%. Go figure.

(Theoretically this would imply that a /logarithmical/ gradient would 
appear "linear" to us; however, that's not exactly the case either, as 
human vision is a bit more complicated than that. My point here is that 
perception of brightness is surprisingly non-linear.)


> You can get the old behavior with assumed_gamma 2.2 or
> by replacing rgb x with rgb pow(x,0.45).

... or by "rgb x gamma 2.2".


> Independent of this issue: Would it be useful to have a
> setting of "color_gamma" that tells POV-Ray how to interpret
> literal color values specified in SDL? The default value of 1.0
> would yield the current behavior while e.g. 2.2 would internally
> convert a value as rgb 186/255 to rgb 0.5 without the need for
> plastering your code with gamma macros.

While the idea is compelling, there's a nasty catch to it: With 
POV-Ray's SDL being as powerful as it is, while lacking a clear 
distinction between colors and vectors, it is pretty difficult - if not 
impossible - to draw a clear boundary between "gamma land" and "linear 
country" that is both safe, sane and self-consistent.

Post a reply to this message