Just want to clarify, that if we have any fine details, especially high-contrast details, Camera AA is the limiting factor/primary thing to set first, because there's no way getting around that to clear up the edges, right? This would suggest to me that setting Camera AA should always be first, and then gauging the impact on other samples afterwards. During testing, we can temporarily lower the AA to figure out what the ideal samples are for diffuse, spec, etc., but ultimately the one thing that's got to be locked for finals is the AA, because that's essentially pass/fail with edges. Yes?
Hi
here is the quote from Cgtalk:
it is practically impossible to get clean shadow with Lights samples = 1, and AA samples =1
so when debugging, Always keep (Camera (AA) =1 and increase Lights samples first, until the shadow are clean
then increase the Diffuse sample by adding "Diffuse_Indirect" AOV, (for diffuse, noise are apparent in "Diffuse_Indirect" ) until you find the best value (8 for example)
then increase the Specular sample by adding "Specular_Indirect" AOV, (for Specular, noise are apparent in "Specular_Indirect" ) until you find the best value(12 for example)
Then move to Transmission, and so on, you get the idea.
each time you find a value make a note for that, and when moving to calculate Specular , you have to set diffuse at 1.and when calculating Transmission, set diffuse and specular to 1.
say you have found these values:
Camera (AA) = 1
Diffuse = 8
Specular = 12
Light = 8
Apply the magic rule: divide /multiply
so diffuse = 8/2 = 4
Specular = 12/2 = 6
Light = 8/2 = 4
Camera (AA) = 1*2 = 2
You can apply the same rule again
Diffuse = 4/2 = 2
Specular = 6/2 = 3
Light = 4/2 = 2
Camera (AA) = 2*2 = 4
Nb: Noise are always stored in indirect_aov.
/Rachid
Read more here http://forums.cgsociety.org/showthread.php?f=87&t=1472960&highlight
This is true, but I think there are different ways to optimize. I actually set my camera rays to 1 to diagnose noise, increasing the sampling on a per-channel/per-light basis. Once I get noise-free results (with jaggy edges), I will then increase camera samples last while proportionally decreasing everything else. Remember,for every camera ray, there are a multiple of channel rays.
As I understand it, the way that raytracing works is that it sends out a bunch of rays from the camera. When the ray intersects geometry, it then sends out additional rays to specifically solve the channel.
So you CAN get clean shading, even if "where" the geometry is has a low accuracy (resulting is jagged edges).
I'm not 100% sure how filtering fits in here.
The simplest way is to set everything to 1 and then increase AA until you have what you want.
If you know what you're doing, then you can start trying to optimize by bringing up other sample settings, like diffuse and specular, while bringing down AA.