I don't understand the huge difference between using +a0.0 and any value
above 0.0.
  Consider this simple scene:

camera { location -z*10 look_at 0 angle 35 }
light_source { <1000,2000,0>, 1 }
plane
{ y,-1 pigment { checker rgb <1,1,.5>, rgb <.8,.4,.2> }
  rotate <-10,30,-30>
}

  When I rendered it with +a0.01 +am2 +r4, the resulting image is the
following:
http://www.cs.tut.fi/~warp/aa1.jpg

  There are big visible artifacts.
  However, when I rendered it with +a0.0 +am2 +r4, the result is almost
perfect:
http://www.cs.tut.fi/~warp/aa2.jpg

  It doesn't seem to matter how small the threshold value is. If it's larger
than 0, it will cause the artifacts.
  I don't understand why.

-- 
#macro M(A,N,D,L)plane{-z,-9pigment{mandel L*9translate N color_map{[0rgb x]
[1rgb 9]}scale<D,D*3D>*1e3}rotate y*A*8}#end M(-3<1.206434.28623>70,7)M(
-1<.7438.1795>1,20)M(1<.77595.13699>30,20)M(3<.75923.07145>80,99)// - Warp -

Post a reply to this message

Warp wrote:
>   I don't understand the huge difference between using +a0.0 and any
>   value
> above 0.0.
The huge difference is: With +a0.0 every pixel is sampled, with any 
value above only pixels which differ to their neighbours are sampled.

>   Consider this simple scene:
> 
> camera { location -z*10 look_at 0 angle 35 }
> light_source { <1000,2000,0>, 1 }
> plane
> { y,-1 pigment { checker rgb <1,1,.5>, rgb <.8,.4,.2> }
>   rotate <-10,30,-30>
> }
> 
>   When I rendered it with +a0.01 +am2 +r4, the resulting image is the
> following:
> http://www.cs.tut.fi/~warp/aa1.jpg
> 
>   There are big visible artifacts.
>   However, when I rendered it with +a0.0 +am2 +r4, the result is
>   almost
> perfect:
> http://www.cs.tut.fi/~warp/aa2.jpg
> 
>   It doesn't seem to matter how small the threshold value is. If it's
>   larger
> than 0, it will cause the artifacts.
>   I don't understand why.
These artifacts appear because it's coincidence (in that image area) 
which color any pixel has and if two pixels differ or not.
If they differ, samples are taken and it looks as it should look. But 
if they don't differ not the mixed color is used for the pixel but one 
of the unmixed colors. So it's coincidence which color any pixel has in 
the final image.
To avoid this problem it's neccesary to take more than one sample per 
pixel to decide if the pixel must be sampled. So the only workaround 
with the current AA methods is to sample every pixel with +a0.0.
Sorry, I don't speak English very well, so please tell me if it was not 
understandable.
Felix Wiemann

Post a reply to this message

Felix Wiemann <Fel### [at] gmxnet> wrote:
> The huge difference is: With +a0.0 every pixel is sampled, with any 
> value above only pixels which differ to their neighbours are sampled.

  Yes, but if the previous pixel was antialiased, its color will differ
from the next pixel no matter where does the ray hit the plane in this
next pixel. The color of the previous pixel is a mix between the light and
dark colors of the plane, and thus is very different from either.
  This would mean that if the pixel at the left of each row is antialiased,
all the pixels in that row should be antialiased as well (because each time
a pixel is antialiased its color will differ radically from the next pixel
regardless of where does the ray hit).
  This is what I don't understand.

-- 
#macro N(D)#if(D>99)cylinder{M()#local D=div(D,104);M().5,2pigment{rgb M()}}
N(D)#end#end#macro M()<mod(D,13)-6mod(div(D,13)8)-3,10>#end blob{
N(11117333955)N(4254934330)N(3900569407)N(7382340)N(3358)N(970)}//  - Warp -

Post a reply to this message

Warp wrote:

> Felix Wiemann <Fel### [at] gmxnet> wrote:
>> The huge difference is: With +a0.0 every pixel is sampled, with any
>> value above only pixels which differ to their neighbours are sampled.
> 
>   Yes, but if the previous pixel was antialiased, its color will differ
> from the next pixel no matter where does the ray hit the plane in this
> next pixel. The color of the previous pixel is a mix between the light and
> dark colors of the plane, and thus is very different from either.
>   This would mean that if the pixel at the left of each row is
>   antialiased,
> all the pixels in that row should be antialiased as well (because each
> time a pixel is antialiased its color will differ radically from the next
> pixel regardless of where does the ray hit).

That's how method 1 works. Method 2 does not care about if the previous 
pixel were supersampled or not, instead it just looks at the 4 corners. My 
method 4/5 does as it also looks at the samples previously taken on the 
pixels borders.

- Micha

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

Post a reply to this message

So ! what is, for most power-user, the best antialiasing setting for the
final render ? I'm still searching for a good method to render the final
image of most of my scenes... (good antialiasing without blur...?!)

Micha Riser <mri### [at] gmxnet> a écrit dans le message :
3caa0067@news.povray.org...
> Warp wrote:
>
> > Felix Wiemann <Fel### [at] gmxnet> wrote:
> >> The huge difference is: With +a0.0 every pixel is sampled, with any
> >> value above only pixels which differ to their neighbours are sampled.
> >
> >   Yes, but if the previous pixel was antialiased, its color will differ
> > from the next pixel no matter where does the ray hit the plane in this
> > next pixel. The color of the previous pixel is a mix between the light
and
> > dark colors of the plane, and thus is very different from either.
> >   This would mean that if the pixel at the left of each row is
> >   antialiased,
> > all the pixels in that row should be antialiased as well (because each
> > time a pixel is antialiased its color will differ radically from the
next
> > pixel regardless of where does the ray hit).
>
> That's how method 1 works. Method 2 does not care about if the previous
> pixel were supersampled or not, instead it just looks at the 4 corners. My
> method 4/5 does as it also looks at the samples previously taken on the
> pixels borders.
>
> - Micha
>
> --
> objects.povworld.org - The POV-Ray Objects Collection
> book.povworld.org    - The POV-Ray Book Project

Post a reply to this message

_Light_Beam_ <fac### [at] aolcom> wrote:
> So ! what is, for most power-user, the best antialiasing setting for the
> final render ? I'm still searching for a good method to render the final
> image of most of my scenes... (good antialiasing without blur...?!)

  The one I used in the second image.
  The bad thing about it is that it's slow.

-- 
#macro M(A,N,D,L)plane{-z,-9pigment{mandel L*9translate N color_map{[0rgb x]
[1rgb 9]}scale<D,D*3D>*1e3}rotate y*A*8}#end M(-3<1.206434.28623>70,7)M(
-1<.7438.1795>1,20)M(1<.77595.13699>30,20)M(3<.75923.07145>80,99)// - Warp -

Post a reply to this message

"_Light_Beam_" <fac### [at] aolcom> wrote:
> So ! what is, for most power-user, the best antialiasing setting for the
                         ^^^^^^^^^^--considering this...
> final render ? I'm still searching for a good method to render the final
> image of most of my scenes... (good antialiasing without blur...?!)

.. you can use: +A0.0 +AM2 +J0.78 +R9 ....

Post a reply to this message

"Warp" <war### [at] tagpovrayorg> wrote:
>   The one I used in the second image.
>   The bad thing about it is that it's slow.

you used +AM2 +R9 ? ... wow ...

Post a reply to this message

Jan Walzer <jan### [at] lzernet> wrote:
> you used +AM2 +R9 ? ... wow ...

  Eh? No. I used +r4.

-- 
#macro M(A,N,D,L)plane{-z,-9pigment{mandel L*9translate N color_map{[0rgb x]
[1rgb 9]}scale<D,D*3D>*1e3}rotate y*A*8}#end M(-3<1.206434.28623>70,7)M(
-1<.7438.1795>1,20)M(1<.77595.13699>30,20)M(3<.75923.07145>80,99)// - Warp -

Post a reply to this message

Jan Walzer <jan### [at] lzernet> wrote:
> .. you can use: +A0.0 +AM2 +J0.78 +R9 ....

  And wait a year for the image to finish?
  I think +r4 should be enough.

-- 
#macro M(A,N,D,L)plane{-z,-9pigment{mandel L*9translate N color_map{[0rgb x]
[1rgb 9]}scale<D,D*3D>*1e3}rotate y*A*8}#end M(-3<1.206434.28623>70,7)M(
-1<.7438.1795>1,20)M(1<.77595.13699>30,20)M(3<.75923.07145>80,99)// - Warp -

Post a reply to this message

Jan Walzer wrote:

> 
> "_Light_Beam_" <fac### [at] aolcom> wrote:
>> So ! what is, for most power-user, the best antialiasing setting for the
>                          ^^^^^^^^^^--considering this...
>> final render ? I'm still searching for a good method to render the final
>> image of most of my scenes... (good antialiasing without blur...?!)
> 
> .. you can use: +A0.0 +AM2 +J0.78 +R9 ....

This causes about 262'000 samples to be taken for each pixel...

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

Post a reply to this message

_Light_Beam_ wrote:
> 
> So ! what is, for most power-user, the best antialiasing setting for the
> final render ? 

I like this:

Render at 300% resolution +am1 +a0.1 +r5 +j20
and bicubic resample to final size.

Warp's test scene rendered 3.5 times faster + 5 seconds
in spent Photoshop. http://luxlab.com/tmp/aa_3x.jpg


_____________
Kari Kivisalo

Post a reply to this message

Kari Kivisalo wrote:

> _Light_Beam_ wrote:
>> 
>> So ! what is, for most power-user, the best antialiasing setting for the
>> final render ?
> 
> I like this:
> 
> Render at 300% resolution +am1 +a0.1 +r5 +j20
> and bicubic resample to final size.
> 

Can you try this scene?:

camera{
location<0,0,-20>
look_at 0
angle 40
}

sphere{0,5 
pigment{
gradient y
pigment_map{
[0 rgb 0.5]
[0.5 rgb .5]
[0.5 rgb 1]
[1 rgb 1]
}
scale .05
translate -10}
rotate z*-33

finish{ambient 1 diffuse 0}



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

Post a reply to this message

"Warp" <war### [at] tagpovrayorg> schrieb im Newsbeitrag news:3caa15d1@news.povray.org...
> Jan Walzer <jan### [at] lzernet> wrote:
> > .. you can use: +A0.0 +AM2 +J0.78 +R9 ....
>
>   And wait a year for the image to finish?
>   I think +r4 should be enough.

probably not enough, if dealing with 16Bit/Chan images ...

... but: Yes, R9 is probably overkill ...

Post a reply to this message

Jan Walzer <jan### [at] lzernet> wrote:
>>   I think +r4 should be enough.

> probably not enough, if dealing with 16Bit/Chan images ...

  I don't understand how the color depth would have any influence in the
need to shoot more rays.

-- 
#macro M(A,N,D,L)plane{-z,-9pigment{mandel L*9translate N color_map{[0rgb x]
[1rgb 9]}scale<D,D*3D>*1e3}rotate y*A*8}#end M(-3<1.206434.28623>70,7)M(
-1<.7438.1795>1,20)M(1<.77595.13699>30,20)M(3<.75923.07145>80,99)// - Warp -

Post a reply to this message

"Warp" <war### [at] tagpovrayorg> wrote:
> >>   I think +r4 should be enough.
> > probably not enough, if dealing with 16Bit/Chan images ...
> I don't understand how the color depth would have any influence in the
> need to shoot more rays.

really ?

If you use +r4 there will be shoot 256 (+AM2) samples per pixel (at max) ...
... this will mean, that this very pixel has a granted accuracy of 1/256 (in
the worst case) ...

... probably enough for a 8Bit/Channel image, as you can't store any more ...

however, in a 16Bit/Channel image, you may want to have the accuracy of at
least 16Bit...

.. this means, however, that you'll need (for the worst case), to shoot 65535
Samples per pixel...
... and this means, you'll need +R8 ...

Post a reply to this message