PDA

View Full Version : Anti-aliasing and the 'look' of the render


ZeroNeuro
05-22-2003, 03:08 PM
I was just sitting here, thinking on the subject of rendering. Specifically, when it comes to the topic of pixel filters, pixel width, and number of samples. I am currently thinking of doing some works for print and for film and was wondering if anyone had any professional tips for what settings are best for this media. Renderman, maya, XSI, whatever, are relevant. So everyone has a chance to get into the discussion.

1) what renderer
2) What filter
3) Number of samples; shading rate where applicable
4) Filter width

Just bored for the moment, and was interested in this end of the process :)

Cheers!
Ritchie L. Roberts
ZeroNeuro Arts Ltd.

joconnell
05-22-2003, 03:51 PM
Okay, the filters I use the most are mitchell-netravelli, catmull-rom, area and video (in max). Mitchell-netravelli seems to give the best overall results for realistic rendering. As regards the number of samples, that changes depending on what I can get away with. I use the least amount of samples possible to get rid of stepped edges / grain. Catmull-rom is great for a really sharp pixar type look but is quite slow and because of the distinct edges mightn't be the most friendly for colour correction. You can get a similar results by rendering with mit-net and then running a sharpen filter in post. Video (like gaussian in mray I suppose) and area I use for more soft cartoony stuff, especially when used with a large filter size. It makes everything look quite fluffy.

ZeroNeuro
05-23-2003, 07:32 AM
Thanks Jo :thumbsup:

Come on, there has to be more interested in this subject. lol

jeremybirn
05-23-2003, 06:06 PM
For film a lot of people render at 1800, resize to 2400 for compositing and output. For print work, the scale-up is sometimes higher, like 200%.

Shading Rate is a Renderman thing, there is no similar control in MR, etc. I usually use a shading rate of 1 for preliminary tests, and go down to .25 or lower for anything final. If there's a lot of hair or little glistening highlights, sometimes .1 is better.

In renderman, where samples don't cost you too much in your render time, I'll usually use 4,4 as a starting point, go up to 8,8 for final. In MR (where more samples multiply up your render time) I usually start with -2,0 for tests, up to 0,2 for final, sometimes as high as 1,3 or 2,4 if there's a lot of detail that could jitter. (The numbers I'm using here are horz,vert for PRMan and min,max for MR.)

-jeremy

Mauritius
05-25-2003, 03:03 AM
Personally, I use a Gaussian 3x3 filter for 2k film work if the stuff will be mixed with live action footage shot on film. If the stuff was shot on HD, I use a 2x2 Catmull-Rom. Pixel samples depend, for previews I use 3x3, as the speed hit isn't that hard for REYES renderers usually. For finals, when motion blur is turned on st. between 6x6 and 10x10 is usually sufficient, even for fast moving objects.

I also feel I have to comment a bit on what Jeremy said:
Mental Ray is a micropolygon renderer, if you want it to be. There are controls for both particularly displacement as well as general tesselation, that generate micropolygon sized triangles. It also deals well with these, memory wise.

Thus you can indeed argue that "shading rate" is not at all only a RMan 'knob'. Only the name is different. However, many other renderers, like e.g. Maya's, have a lot of procedural textures which are not antialiased very well. Even the Maya 5 docs still list the warning about using the "Min Screen" tesselation control with displacement shaders as it will likely result in popping of high frequencies as object size in screen space changes. This renders the option unuseable in many situations.

Regarding shading rate: I usually test with 10 or larger and do final renders with 1. If you have to go below 1 for the final render in an average scene, this usually is a sign that your shaders need some antialiasing treatment or you should consider using some LODs for sudden geometry.
Hair/fur/etc. is another issue, but what many RMan users miss at first is the fact that the width of a curve primitive already puts an upper bound on micropolygon size in the curve's u direction, which is completely independent of the active shading rate.
And of course, shading rate is an attribute and not an option -- it should be set per object, not globally, per scene. This allows for very large shading rates on curves, unless you need shading variations across the curve's width, with a frequency higher than 1 in the u direction of uv space. Using smooth interpolation usually cuts it in such a case. Shading rates like 20 or the like are no problem, unless your hair is really very curly.
What has rather to be high, when rendering thin curves, are the pixel samples. Using st. like 20x20 if motion blur is turned on is no exception. Worth mentioning is that motion factor can become a problem with large shading rates -- particularly on curves -- and lead to highlight popping/flickering.


Cheers,

Moritz

CGTalk Moderation
01-15-2006, 05:00 AM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.