The Second Antialiasing Pass (Curiosity)

[Rodney]

This is yet another one of those general topics where I'm trying to formulate the question but don't have enough to get that done.

 

I note that in many of my renderings the initial rendering passes that A:M reports tend to render pretty quick (note that this is with and without multipass on) even on the order of seconds.

The initial antialiasing pass can take about that same time.

During the antialiasing second pass things slow down considerably (a render last night took seconds for each of the initial passes but over an hour for the second antialiasing pass).

 

Now granted... there may be a lot going on there in that pass and some of it might might not specifically be related to actual antialiasing... although that's what it states is happening.

Also... roughness of surfaces and shiny reflections surely added considerably to that particularly lengthy pass.

I have no issues with that.

 

But my curiosity is officially aroused and I want to learn more about how best to wrangle this second antialiasing pass and to better understand how to manipulate it.

 

Any insights, pointers or references will be appreciated.

 

 

Also as an aside: Over the last two updates I distinctly detect better renderings in A:M although I am not aware of any specific changes. It could be and very likely is just me and my aging eyes but I like what I'm seeing these days. And regardless, A:M still amazes me.

[robcat2075]

I can recall an answer I gave to a similar inquiry...

The dig difference between regular and multipass is how the anti-aliasing is done.

Multi-pass subdivides each pixel (16 passes would be a 4 x 4 subdivision) and does a full render for each sub pixel then averages the sub pixels together.

However, a pixel that is among other pixels of similar color (like in a blue sky) needs no anti-aliasing and won't benefit from it. Multipass rendering wastes time subdividing it.

Regular render does "adaptive anti-aliasing." Every pixel is rendered, then the contrast between pairs of pixels is compared. If the contrast is above some threshold the pixels are subdivided and re-rendered on a 2x2 grid. The sub pixels are again compared for contrast and if any pairs are above the threshold they are subdivided again and rendered.

The result is that regular rendering has anti-aliasing detail equivalent to 16-pass rendering, but usually in far less time.

It is possible for an extremely detailed scene to take longer in regular rendering and I have noticed that objects with thin lines look better with multipass.

So, the difference between the first AA pass and the second AA pass can be as much as the difference between rendering 4 sub pixels and rendering 16 sub pixels (worst case).

[Rodney]

You know... I even found that topic via search before but I didn't catch the part where you emphasize what is occurring while NOT using multipass.

Since I thought it pertained to Multipass I moved on too quickly.

 

Hmmm... yes.

Something to chew on.

 

 

Thanks Robert!

 

 

 

 

[robcat2075]

 

 

Also as an aside: Over the last two updates I distinctly detect better renderings in A:M although I am not aware of any specific changes. It could be and very likely is just me and my aging eyes but I like what I'm seeing these days. And regardless, A:M still amazes me.

 

 

If you have an example of that, it would be interesting to see.

[Rodney]

If you have an example of that, it would be interesting to see.

As per usual I don't but my sense is that images are crisper if not slightly better lit (with minimal changes from the defaults that is)

[robcat2075]

This is yet another one of those general topics where I'm trying to formulate the question but don't have enough to get that done.

 

I note that in many of my renderings the initial rendering passes that A:M reports tend to render pretty quick (note that this is with and without multipass on) even on the order of seconds.

The initial antialiasing pass can take about that same time.

During the antialiasing second pass things slow down considerably (a render last night took seconds for each of the initial passes but over an hour for the second antialiasing pass).

 

 

This might well be a case where multipass gets you a faster result with comparable quality.

 

There is no waiting for an anti-aliasing pass in multipass, you just have to wait for the multiple passes.

 

 

If you had a long animation to render, i'd test a few representative frames in both modes, see which is preferred and go with that for the whole run..