It's really just a function of the sampling.  If you recall the
particular look of atmosphere, it had a distinct banded look.  Even
though it was way off, it tended to look smoother because it was so in
at least on direction.  But that was just because it sampled at equal
lengths along the viewing ray.   Using jitter gave it a look similiar to
media.

The grainyness is objectionable though.  Something I've been thinking
about is that there must be some way to smooth media without taking
extra samples.  I figure it should involve comparing pixels and trying
to balance them.  I suppose a new keyword could let the user control
this by allowing the user to give a maximum intensity difference that
the pixels should be smoothed.  The maximum for 24 bit images should be
1/255, but higher values would likely be specified.  Perhaps the sampled
pixels could be grouped into fours and cached, then they could be
compared against each other.  Then the intensities would be adjusted
until neighboring pixels don't vary by more than the specified amount.  

I've heard that similiar techniques are used for stachastic renderers
for the same reason.  There is a certain loss of accuracy, but it
doesn't require much extra time and results in more visually pleasing
images.  If you take that image and draw a freehand mask around it in a
paint program, then apply a gaussian blur to the image, I bet the
results would be very nice.

comments/flames?

-Mike

Ken wrote:

>   I too will avoid comment on the specifics of this scene which is
> note worthy in and of itself but one thing concerns me about this
> image and other recent image postings using the new media feature.
>   That concern is the grain structure that seems to be an inherent
> artifact of the process. While not as pronounced in the smoke of
> the above mentioned image, probably due to it's distance, it really
> shows up often in many other images. I wonder if this is going to
> be a fact of life with this feature or if there is a something that can
> be done to reduce the particle size while maintaining the density
> needed for realism.
>   Smoke paticles are very fine particulate matter and the reason it
> appears to us as it does is primarily due to shear volume. Might this
> grain artifact be a problem of image resolutions and screen capabilities
> or am I just jumping the gun on this still evolving feature ?
> 
>   I'm willing to entertain discussion on this but truthfully have as yet
> not spent much time with it and am by no means a barometer of
> it's potential.
> 
> --
> Ken Tyler
> 
> tyl### [at] pacbellnet

Post a reply to this message