Warp wrote:
>   Even though the idea and theory behind the new antialiasing methods
>   sound
> great, if I'm completely honest, the example pictures are not really
> convincing. Although they do a somewhat better job in many cases, they
> still don't do it so much better.

The new methods are thought as replacement/improvement of method 2. If you 
compare method 4 to 2 in the examples 4 does in most scenes considerably 
better. Especially the thin lines are much better! With that scene it does 
even noticeable better than method 1 when you look at the lines in the 
middle of the picture.

Compared method 4 to 1 it is true that there is not much improvement 
visible though I can hardly objectively measure what is 'better'.

>   The test scene 6 is a perfect example of what I'm talking about. Even
> though the description of the new methods sound like they would make the
> image a lot better, in practice they don't perform radically better than
> the existing methods in practice. The graininess is still there, and I
> don't really see any considerable improvement with the new methods.
> 

Hmm, I really do not see much graininess in the test scene 6, test II, 
method 4 left.. I'd say it is less than with method 1. I had limited the 
test at 3 spp because I considered the results as 'close to the perfect'. 
If there is still to much graininess for you, you can try lowering the AA 
threshold with method 4 just a bit and you'll see that it gets better at 
the critical places.

I would sum up the test results like that:

1 (concentrical circles): Method 4 needs fewer samples as both 1/2 to get a 
nice result. Method 1 would need a huge number of samples to get a perfect 
result.

2 (thin lines): Method 2 fails compleatly (You would need a threshold of 0 
to get acceptable results). Method 1 gives good result at low settings but 
at high settings the lines are not smooth in the middle. Method 4 soon 
gives smooth lines on the left side and can compete with method 2 on the 
right.

3 (small dots): Only Method 5 gives acceptable results.

4 (checker plane): Up to the distance where individual squares are 
distinguisable method 4 does well. Behind, method 1 does besser. Best would 
probably be to blur things from that distance on. Method 2 gives 
catastrophical results.

5 (bricks): I cannot really conclude from this scene. Even the 'perfect' 
scene has visible structures. Method 4/5 certainly do better than 2. 

6 (real scene): IMO the graininess disappears faster with method 4 than 
with the other methods. 

- Micha

-- 
objects.povworld.org - The POV-Ray Objects Collection
book.povworld.org    - The POV-Ray Book Project

Post a reply to this message