PDA

View Full Version : Why doesn't adaptive AA like me?



xxiii
11-01-2011, 07:37 PM
Attached, the same scene rendered with AA set to 64, and with Adaptive AA. No matter what I do, I can't get the Adaptive version to look like the 64 version (or any better than it looks in this attachment). At the most, Adaptive seems to do two extra passes in the blurry parts of the image, when it needs to do several dozen. (in manual settings, 16 still looks graney, 32 starts to look good)

This seems to defeat the entire purpose of Adaptive, which is supposed to target the areas that need it the most, and have at it.

Lewis
11-02-2011, 01:41 AM
Can you post scene ?

did you try to turn on Oversample at 0.7 ?

cptwhite
11-02-2011, 03:39 AM
You should have a read of this (http://www.except.nl/lightwave/aa/index.htm)

It's a fantastic guide which runs through what all the various settings do, typical range values you should be looking at, and what the "classic" camera is most similar too.

You will see a table where it shows you the AA value and the maximum number of adaptive sampling passes you will get at that AA level. It is stepped and as the AA goes up the maximum passes of AS goes down (which is logical - you need less fine correction the greater the overall AA level is).

Threshold value is the contrast difference between 2 adjacent pixels (i.e lower value is more critical and more adaptive sampling will occur giving better AA results)

Oversampling can be likened to "selection feather value" or "blur" in photoshop, where the value extends the AA sampling beyond the usual sample range to give a localised "blurring" or "softening" of the adaptive sampling. As I understand it the value represents the value it extends in pixels (so 0.5 would be half a pixel for example)

You need to confirm what your current AA setting is (when using adaptive sampling), but to hazard a guess, based on the fact you're using 64 passes for the standard benchmark, try something like:

AA = 16
Threshold = 0.03
Oversampling = 0.5 (or higher if it's still pixelated)

Let us know how you get on.

Danner
11-02-2011, 03:42 AM
what is the adaptive sampling theshold ?

erikals
11-02-2011, 07:25 AM
try this,...

AA=3
AS=0.1 (or even 0.05) lower value makes better quality...

xxiii
11-02-2011, 01:39 PM
Here is the scene, (except I'd changed the floor to white in my initial post).

Above a certain point with AA, it acts like its not going to make any additional (adaptive) passes, no matter what. As saved here, threshold is set to the minimum it will go (without hitting 0). I also tried the various suggestions made so far.

I find myself wishing some of these settings (such as max passes) were directly settable instead of inferred, but in this case, as far as I can tell, threshold just isn't working, or else its calcuation of how many additional passes it should do isn't working (which seems to be the case based on the progress bar).

I might give "hank the donut" a try and see what happens, but it seems like something just ins't right in the Adaptive system. I would think a threshold of .0001 would pretty much force it to use the maximum samples on the entire scene, which according to the except guide should be 4 more passes for an AA of 16, but in watching the progress bar, one can see that it doesn't even allow space for these potential extra passes (whereas with lower AA, the progress bar DOES allow room and you can see the 1 or two extra passes it makes).

Lastly (this scene is about 4 years old) is there a compatibility switch I need to turn off somewhere that is perhaps making it act this way?

I also had a thought that perhaps it was calculating aliasing before doing color-space correction, (I've tried it with and without) but it still looks pixelated either way, and still refuses to do additional adaptive passes.

erikals
11-02-2011, 02:09 PM
not sure why, but try this,

AA = 99
AS = (off)
OS = (off)

3DGFXStudios
11-02-2011, 02:19 PM
Why are you using AS if you're using DOF? It's faster to turn it of and your DOF will look nicer ;)

Lewis
11-02-2011, 02:41 PM
Hows this with proper AA/AS settings in 78 seconds :)?

AA 2, AS 0.005, OS 0.7 and few small tweaks ;).

Lewis
11-02-2011, 03:12 PM
BTW here is scene file where AS kicks in very much. Actually you could turn down AA even to 1 if you wish, it's little more grain but not much more and it'll shave of another 6-7 seconds of rendertime ;).

JonW
11-02-2011, 06:17 PM
First:
AA 513 passes: 4:09

Second:
AA 2
AS 0.03
OS 0.7
Motion Blur Photoreal 5: 24.5 seconds

Motion Blur of 3 is ok: 14.5 seconds

erikals
11-02-2011, 06:34 PM
nice...! fast..!

JonW
11-02-2011, 07:14 PM
Now it's getting silly!

Here is AA 4097 passes: 35:21

I've found after 600 to 700 passes things start to go wrong.

xxiii
11-02-2011, 10:35 PM
Hows this with proper AA/AS settings in 78 seconds :)?

AA 2, AS 0.005, OS 0.7 and few small tweaks ;).

Ok, 68.6 seconds for me :D

Playing around with the differences, it appears "Use Global" in the motion options is the culprit, even though my global motion options are set to exactly the same thing.

This is similar to the bug I originally created this scene to investigate/demonstrate, which I thought was fixed now. This used to occur when the "camera use global" was checked, and if I remember right, motion options didn't used to have its own "use global" but I'm fuzzy on this.

When I resurrected this scene, it didn't occur to me to (re)check the use global settings (except for the camera one which I did try, having been the cause of this problem last time) when the old problem resurfaced.

So, do others agree this seems to be a bug, and I should (re)report it? Or, is there some logical (I fail to see how) reason it should be behaving this way?

xxiii
11-02-2011, 10:43 PM
Why are you using AS if you're using DOF? It's faster to turn it of and your DOF will look nicer ;)

I'm not sure why DOF and AS should be mutually exclusive, or faster or better looking with it off? But if this same (apparent) bug has been biting you as well, perhaps it lead to this conclusion?

Lewis
11-03-2011, 01:43 AM
Yes i've turned off globals and changed sampling mode and pattern but i think Globals is the problem here for some reason. I never use globals so i wouldn't run into that problem on my side if i didn't try your scene ;).

JonW
11-03-2011, 07:14 AM
A few more trial & errors: AA 70, 80, 90, 99, 100, 110, 115, 117, 118, 120, 129, 200, 240, 250, 270, 280, 300, 350, 400, 450, 480, 500, 513, 529, 576, 588, 600, 610, 625, 626, 630, 650, 700, 769, 800, 4097, 25000 (crop)

AA 513 is actually very poor quality.

AA 100, 200, 300, 400, 500, 600 are all good with increasing quality & easy to remember. 400 is about 99% of the way there, 626 is as good as it gets. Then it's down hill from there.

Attached AA 400, 626, 25000/626

Andy Meyer
11-03-2011, 09:05 AM
you need to set AA >1 in most cases.
AS works on the result of the AA pass. AA = 1 can hide things from AS (fine lines, noise).
if you use other than linear color space AS need extreme low value = extreme long render times. sometimes AA only is fast with sRGB color space if you have dark areas or details. AA/AS can solve noise, but you should set up GI and shows to not make too much noise.

xxiii
11-03-2011, 01:24 PM
I reported the issue with the "Use Globals" checkbox. In what rivals the fastest turn-around for a commercial bug report I've ever had, it has been acknowledged as having apparently been dealt with.

I now lump it in with several other bug reports that I'm supposed to re-test when I can (so I can get rid of the "apparently" above).

littlewaves
11-03-2011, 03:57 PM
Why are you using AS if you're using DOF? It's faster to turn it of and your DOF will look nicer ;)

really?

I've always experienced quite the opposite.

Low AA and .03 AS has always give me far better and faster results than trying to crank up the AA on its own to smooth the blur

Elmar Moelzer
11-03-2011, 04:17 PM
Why are you using AS if you're using DOF? It's faster to turn it of and your DOF will look nicer
Nope, it wont.