"nemesis" <nam### [at] gmailcom> wrote:
> "arblick spule" <aspule> wrote:
> > How do?
> >
> > Though I'd chuck my two-pence into the ambient occlusion struggle with POV.
> >
> > After looking and playing with radiosity solutions for AO, I found the results
> > to be lacking something.  It didn't really square up to the proper AO offered by
> > the more expensive renderers.
>
> AO is nothing more than a fake GI method.  It doesn't take light into
> consideration, only how far geometry is from each other to affect occlusion.
> This way, it might occlude in shadow some parts even if it was entirely under
> direct light.  That means it's truly fake and does not correspond to reality.
>


Indeed, however GI is fake as well.  Let us not forget that math doesn't always
equal real(istic).

All images that come out of all renderers (freeware or otherwise) have
noticeable flaws in them.  For instance the dragon render above, without AO
pass, has a very bright hard-palate despite all three of the lights being
"behind" it and the material having quite a low reflectance value.  Radiosity
was turned on for this render and yielded a, quite frankly, wildly displeasing
picture.  3DSMax's renderer will do exactly the same and I have no doubt that
most of those that exist will too.  What, I opine, AO does is give an obviously
fake image (Dragon sat on checkered disc) some degree of - well - realism for
the eye, or maybe "depth" is a better word.

Hang on.  Ambient occlusion is generated by firing off a load rays from random
points towards the pixels 3D/2D intersection.  If the object/pixel doesn't get
hit directly then it is considered occluded, but only from that angle.  AO takes
a few (generally 40 or so) samples and averages them to get an "occlusion level"
if you like.  At least this is how Mental Ray describes how it does it.  So it
differs from global illumination by not bouncing light from object to object,
but assessing how much ambient light "could" hit the target even if there was no
light at all.  'Tis all theoretical.

The problem I guess here is differing artistic acceptability (if that makes any
sense...). I will trade off technical realism to have an image that looks nice,
to put it simply.


Original render:
    http://i53.photobucket.com/albums/g69/aspule/BoxesNoOcc.jpg

Ambient occlusion pass:
    http://i53.photobucket.com/albums/g69/aspule/BoxesOccPass.jpg

Final image:
    http://i53.photobucket.com/albums/g69/aspule/BoxesOcc.jpg

This is where my method falls down!  Look at the shadows around the base of the
closest box.  Not very good!  This could be corrected by rotating the average
angle towards the real light source, I guess...

.....Hey ho...              ...Peace out!

Post a reply to this message