The supersampling anti-aliasing(SSAA) that we learned in this lecture happens to be the slowest one but has a great output.

Besides SSAA, we have Multisample anti-aliasing(MSAA), which is an improved version of SSAA. It reduces duplicated calculation of shading process for each sample point. Instead, it does the shading calculation once per pixel and multiply it with the ratio of "how much of the pixel is covered".

MSAA has a drawback. It doesn't fit the deferred rendering frame work very well. Thus we have some Post-proccessing-based methods like FXAA and TXAA.

JCDenton

FXAA is pretty interesting - it doesn't actually analyze the primitives - it merely performs an edge detection then performs smoothing on a per pixel approach.

tiffany2

I was wondering in class how expensive supersampling was. Is this something that is parallelizable?

WJM

@tiffany2 I found this paper that goes into parallelizing super-sampling, specifically using the method JCDenton suggested of only performing on all edge pixels. https://ieeexplore.ieee.org/document/651518

There are a lot of methods of anti-aliasing.

The supersampling anti-aliasing(SSAA) that we learned in this lecture happens to be the slowest one but has a great output.

Besides SSAA, we have Multisample anti-aliasing(MSAA), which is an improved version of SSAA. It reduces duplicated calculation of shading process for each sample point. Instead, it does the shading calculation once per pixel and multiply it with the ratio of "how much of the pixel is covered".

MSAA has a drawback. It doesn't fit the deferred rendering frame work very well. Thus we have some Post-proccessing-based methods like FXAA and TXAA.

FXAA is pretty interesting - it doesn't actually analyze the primitives - it merely performs an edge detection then performs smoothing on a per pixel approach.

I was wondering in class how expensive supersampling was. Is this something that is parallelizable?

@tiffany2 I found this paper that goes into parallelizing super-sampling, specifically using the method JCDenton suggested of only performing on all edge pixels. https://ieeexplore.ieee.org/document/651518