Diving into Anti-aliasing

The link: http://www.beyond3d.com/content/articles/122

 

Quick heads up: figure 3 is the same as figure 1 (the wrong images are used).

 

The correct images are:

http://www.beyond3d.com/images/articles/diveintoaa/illus/8x8pixelgrid_lines.png
http://www.beyond3d.com/images/articles/diveintoaa/illus/8x8pixelgrid_lines_rasterized.png
http://www.beyond3d.com/images/articles/diveintoaa/illus/8x8pixelgrid_lines_perfect.png

 

Nice. Thanks for that, Peter.

 

Oh crap, I read it at least 3 times in the preview and always missed that. Thanks for the heads-up.

Thanks, my pleasure (yeah, I actually had fun writing that article).

Edit:
Just saw my join date (I haven't actively posted here in quite a while), and yeah, the topic is even older than that

 

Cool! Nice, thanks a lot!

 

Great article, thanks a lot! Will be used as a reference for a long time.

 

I have a noob question
whats the difference between downsampling and supersampling the article refers to them differently

 

Normally downsampling means rendering at higher resolution and shrink it with some filter (generally a box filter) to a lower resolution. So it's equivalent to use a grid sampling positions. Supersampling simply means a pixel has multiple subsamples, which does not have to be in grid positions.

 

Like pcchen said in this context it basically refers to the difference between ordered grid and sparse grid SSAA.

 

Great article PeterT, I really enjoyed it .

It was super nice reading a front page article again.

 

I've only just glanced at a couple of pages but it appears to use a box filter for the samples which, at least for a reference result, is, IMHO, flawed.

[Update] Got to page 7:

I'm not sure I'd entirely agree with that. A tent filter, although a massive improvement on a box, still has poor high frequency response. Better filters can be used which could maintain 'sharpness' while eliminating more aliasing.

 

I am hoping for more elaboration on this point. the way CSAA works always seemed fuzzy to me.

Otherwise this is an excellent article.

That's not the way I understand it, and I am open for corrections.

Both Down/Super Sampling can use Ordered or Sparse Grids .. these are merely positions to take the sample .

SuperSampling is just a technique that treats each sample position as a true pixel, performing all the usual computations on it. It can be done directly on the output resolution. (ie, at 720p on a 720p monitor).

DownSampling however is not. it is achieved through rendering the game at a resolution that is higher than the output one (ie, at 1440p on a 720p monitor), then shrinking the image with a combination of Grid/Box/Tent methodology.

 

Normally when one talks about "rendering at a higher resolution" it generally means rendering into a larger rectangle with more pixels. Therefore, the sampling positions must be a grid (unless you have some sort of very weird pixel arrangement for the high resolution rendering, but that's not really possible with current hardwares).

 

Went through it once, seemed good, lacking in details when it comes to AA, also did I miss the explanation about post processing so called "AA" being smart blurs ? (No extra sample = not anti aliasing to me)

 

I reply to the aliasing subject before I read the article:I feel aliasing is a product of hardware .I mean,a lot of things have to be timed perfectly for HD to occur!oh there will still be a bit of aliasing but it will be at the pixel level (probably badly visible)when was the last time people tried to fix aliasing from the hardware side?

From my experience ,color profile is often the culprit of non-hd looking image !now a day aliasing coming out of Dev is barely visible,so along the way from Dev to gamer something change!

 

On page 6 :about real time calculation of super sampling aa!wouldn't the 2014 (newest)method of sparse fast Fourier transform be of huge help ?

 

About post possessing?since this is a dynamic analytical tool?wouldn't this have nefarious indirect interaction with one feature color profile of all kind supply?(namely perceptual or relative!)does this mean if I use a post process (like AMD mlaa)I need to adopt absolute colorimetry?if this is the case :how the eck do I do this?game maker tend to keep this under lock and key and react badly if user manually bypass !