|
there's no best setting.
box 1/1 is the fastest I think, but not the best
for sure.
triangle 1/1 is for crisp renders, no added
softness.
gauss 3/3 is good for a soft look, especially with
hair/fur rendering and the like, or when doing
clouds or so.
mitchell 4/4 is good for high quality,
sharp rendering with still good antialiasing. its also the slowest of the
ones mentioned.
I tend to use the same settings for all passes that
have the same content, in order to avoid edge problems.
so if I do the beauty with mitchell 4/4,
I'll do matte or holdout passes or so with mitchell too, even if it is quite
expensive.
----- Original Message -----
Sent: Thursday, February 22, 2007 7:47
PM
Subject: Re: 16bit Zdepth?
You are all going to be dreaming of samples tonight if this
thread continues....ugh. Look, I just want to know one thing.......what
setting makes my render look best :-)
André
Adam wrote:
The
if(allsamples=0){} idea is a good explanation why both mechanisms seem to
work. Thanks for the input! :)
-André
Halfdan Ingvarsson wrote:
A) Triangle filter of size 1 will have zero
weights on the edges. So in the case of 1/1 samples, the middle sample
will have a weight of 1.0 while the rest (which are all on the edges of
the pixel) will have a weight of 0.0. Therefore the middle pixel is the
only one sampled.
B) I'm guessing from the couple of tests that I
did that in the case of a combined weight of zero over all samples covered
by the filter, mental ray will simply pick a single, unweighted sample to
use. Possibly, the lower-left pixel sample. In this case the image shifts,
compared to the center-sample case.
The filter is always
rectangular, not elliptical.
- ½
-----Original
Message-----
From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM]
On Behalf Of André Adam
Sent: 22-Feb-07 12:45
To: XSI(at)Softimage.COM
Subject: Re:
16bit Zdepth?
That doesn't seem to work for my logic; what happens
when rendering Sampling 0/0 with a Triangle or Gauss Filter set to 1?
A) If the filter size is a radius in pixel length, and the filter
works from the pixel's center rather from than the corners, then four
samples will be hit by the filter. Five in case of a Sampling 1/1. Should
therefore show an aa effect.
B) If the filter size is a diameter
in pixel length, and the filter works from the pixel's center rather than
from the corners, while four samples are hit by the filter, all of their
values are multplied by a zero weight and the result should be a pitch
black picture. Certainly doesn't work this way.
Looking at the
empirics, I get perfectly non-aa pictures with both techniques, sampling
1/1 and sampling 0/0 combined with either triangle or gaussian filter set
to 1. Though the pictures differ a bit.
What's the missing piece
in the puzzle now? :)
Cheers!
-André
Halfdan Ingvarsson wrote:
Mental ray samples on pixel corners but filters
from the pixel centre.
With that in mind, the only way to get a
non-shifted, non-blurred, point-sampled image is to render with sampling
at 1/1 and using a triangle filter at 1. This ensures that only the
center sample is used in the final output, with the rest thrown out.
Unfortunately, it is also hideously expensive.
- ½
-----Original Message-----
From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM]
On Behalf Of André Adam
Sent: 22-Feb-07 11:54
To: XSI(at)Softimage.COM
Subject: Re:
16bit Zdepth?
The minimum filter range is 1 pixel (radius), and
since the samples are shot in a one-sample-per-pixel raster, the filter
will *always* reach the neighbouring samples. The trick is to use a
filter with a zero-weight at the maximum extent, like triangle or gauss.
Box will have a 1 weight throughout the whole distance, therefore
combining the original pixel's sample value with the neighbouring ones.
Oh, and I guess the jitter option should be turned off for all
of this to work; haven't tried that, but it should cause problems due to
the need for an even sampling raster (see above).
Cheers!
-André
Daniel Rind wrote:
It's Sampling at 0 / 0, and filter size at 1
(the minimum).
Each pixel is sampled at least 2^2/min/
times and at most 2^2/max/ times in each direction.
So
for 0 / 0 this means 1 x 1, and for 1 / 1 it would mean 4 x 4, or 16
samples per pixel.
The filter /shouldn't/ make any
difference, but oddly it does. Box filtering adds blur and shifts the
whole image down and left by half a pixel (really odd), so take
triangle filtering.
Ciao, Daniel!
2007/2/22, Tim Leydecker <BauerOink(at)gmx.de <mailto:BauerOink(at)gmx.de>>:
Hey,
thanks alot
guys! I´ve had great success with a projection
of a gradient piped into a constant shader and that one
assigned the partition (like it simple, the
backroundpartition).
It´s just that
rendering bit that puzzles me. To rinse and repeat,
I use Filtering set to 0/0, or is it 1/1?
That´s the bit I don´t get, how to make
sure I have no AA at all.
From there it´s
still somewhat challenging to get the correct
mapping (e.g. linear instead of gamma corrected) out of XSI
but that one I should get together by rendering
to float/linear.
I can´t test the above
today since I´m still working in Maya
but
allready do most of my personal stuff in XSI (at home) it
would be great to also optimize that in depth
once and for all.
The more I have the
chance to compare the apps, the less
satisfied
I am with Maya, btw. Especially when shading and
rendering with passes in mind. Oh, boy.
Additionally, it´s Maya8...
Cheers
tim
-----
Original Message -----
From: "Joe Laffey"
<joe(at)laffey.tv <mailto:joe(at)laffey.tv> >
To: <XSI(at)Softimage.COM <mailto:XSI(at)Softimage.COM>>
Sent: Thursday, February 22, 2007 4:18 PM
Subject: RE: 16bit Zdepth?
> On Thu, 22 Feb 2007, Kim Aldis
wrote:
>
>>>
-----Original Message-----
>>> From:
owner-xsi(at)Softimage.COM <mailto:owner-xsi(at)Softimage.COM>
[mailto:owner-xsi(at)Softimage.COM
<mailto:owner-xsi(at)Softimage.COM>]
On
>>> Behalf Of Andreas Bystrom
>>> Sent: 22 February 2007 13:25
>>> To: XSI(at)Softimage.COM <mailto:XSI(at)Softimage.COM>
>>> Subject: Re: 16bit Zdepth?
>>>
>>> grab a scalar-state, plug that into a change_range node
and
finally
>>>
plug the change_range into the surface port, leave the
scalar-state
>>> settings as it is and set the "new range start" to
0 and "new
range
>>> end" to 1 in the change range node, your old and
new-range
parameters
>>> should be set to the distance where you want the
depth-pass to
start
>>> and end, you can link these 2 values to nulls with an
_expression_ and
>>> use those to control the depth-pass
>
>
>> You may also find that a linear
falloff isn't necessarily what
you need.
>> Plugging the output of the change
range into a grad will allow
you to tweak
>> it, if you need to.
>>
>
> I think the non-linear mapping would be
better done in the
compositor, though. That
> way you have the option of linear or
putting a curve on it.
>
> --
> Joe
Laffey
| Visual Effects for Film and
Video
> LAFFEY Computer Imaging
|
-------------------------------------
> St.
Louis,
MO
| Show Reel http://LAFFEY.tv/?e04685
>
USA
|
-------------------------------------
>
.
| -*- Digital Fusion Plugins
-*-
>
--------------------------------------------------------------------------
> Mail here will be rejected --> "Real
Trap" <
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the
following text in body:
unsubscribe xsi
|
