Sub-pixel anti aliasing

[MistaPi]

Just a quick question.
What does sub-pixel anti aliasing mean? Dosent both supersampling and multisampling use sub-pixels to perform anti aliasing?

 

[3dcgi]

Without knowing the context of the statement I can't answer the question for sure, but you're correct. Saying sub-pixel antialiasing is a little redundant. Super and multi-sampling both render to a sub-pixel level and average the sub-pixels to get the final pixel.

 

[Crusher]

It sounds to me like it implies using multiple samples from one or more pixels, as opposed to just taking one sample from each of several pixels. That is, Sub-Pixel Antialiasing is not redundant, just a broader generalization of the type of technique multi- and super-sampling use. Sub-Pixel Super-Sampling or Sub-Pixel Multi-Sampling would be redundant.

 

[KimB]

Sub-pixel anti-aliasing doesn't sound that way to me. Sub-pixel carries with it the connotation "smaller than a pixel." All current anti-aliasing techniques use many samples smaller than the final pixel output (At least for z-buffer testing, if not for framebuffer writes).

 

[Crusher]

Essentially, "multiple samples from one or more pixels" = "many samples smaller than the final pixel output", since the pixel I was referring to was the final output pixel. I guess to be completely accurate, I should have said "multiple samples from the same area that is covered by one final pixel", but I didn't realize I had to be so specific in order to point out the difference between that and simply averaging pixels by taking one sample from each at the level they're being displayed at. It's marketing fluff any way you cut it, since nobody does "blur-aliasing" anyway.

 

[Hyp-X]

Sub-pixel anti-aliasing doesn't sound that way to me. Sub-pixel carries with it the connotation "smaller than a pixel." All current anti-aliasing techniques use many samples smaller than the final pixel output (At least for z-buffer testing, if not for framebuffer writes).

Ahem.

Samples are points. They don't have a size (extent).

So samples are not smaller, they are only closer to each other.

 

[KimB]

In a manner of speaking. The samples in the framebuffer that exist before downsampling can be thought of as smaller than the output pixels. You can also consider the 'size' of a pixel as how much effect it has on the screen output. But we are splitting hairs here.