User Manual

Anti-Aliasing of Aux Channels in the OpenGL Renderer
The reason Fusion supplies separate anti-aliasing options for color and aux channels in the
Anti-Aliasing preset is that supersampling of color channels is quite a bit slower than aux
channels. You may find that 1 x 3 LowQ/HiQ Rate is sufficient for color, but for world position or
Z, you may require 4 x 12 to get adequate results. The reasons color anti-aliasing is slower are
that the shaders for RGBA can be 10x to even 100x or 1000x more complex, and color is
rendered with sorting enabled, while aux channels get rendered using the much faster
Z-buffer method.
TIP: For some things, sometimes using an SS Z-buffer improves quality, but for other
things like using the merge’s PerformDepthMerge option, it may make things worse.
Do not mistake anti-aliasing with improved quality. Anti-aliasing an aux channel does not mean
it’s better quality. In fact, anti-aliasing an aux channel in many cases can make the results much
worse. The only aux channels we recommend you enable anti-aliasing on are
WorldCoord and Z.
TIP: We strongly recommend disabling Anti-Aliasing on Material ID and Object ID
channels, TexCoord, Normal, BackVector, and Vector channels. The issue arises when
you have multiple 3D surfaces with radically different TexCoord values in one pixel.
The anti-aliasing does not restrict itself to sampling the main surface but samples both
surfaces. For example, if one surface has TexCoords that are approximately (u,v) = (0,
0) within that pixel, and the other surface has (0.5, 0.5), you get a blending of these
two. The blended area of the texture could have colors like (0, 0) or (0.5, 0.5), resulting
in an oddly colored pixel artifact being output from the 2D Texture node. The same
problem can happen for normals.
Enable (LowQ/HiQ)
These two check boxes are used to enable anti aliasing of the rendered image.
Supersampling LowQ/HiQ Rate
The LowQ and HiQ rate tells the OpenGL render how large to scale the image. For example, if
the rate is set to 4 and the OpenGL renderer is set to output a 1920 x 1080 image, internally a
7680 x 4320 image is rendered and then scaled back to produce the target image. Set the
multiplier higher to get better edge anti-aliasing at the expense of render time. Typically 8 x 8
supersampling (64 samples per pixel) is sufficient to reduce most aliasing artifacts.
The rate doesn’t exactly define the number of samples done per destination pixel; the width of
the reconstruction filter used may also have an impact.
Filter Type
When downsampling the supersized image, the surrounding pixels around a given pixel are
often used to give a more realistic result. There are various filters available for combining these
pixels. More complex filters can give better results but are usually slower to calculate. The best
filter for the job often depends on the amount of scaling and on the contents of the image itself.
The functions of these filters are shown in the image above. From left to right these are:
Chapter – 80 3D Nodes 1696