Page 1 of 1
Multipass feature
Posted: Fri Aug 20, 2010 12:58 pm
by rman1974
Is it possible to implement some kind of Multipass feature - when octane would render several passes with different seed and then merge them together to reduce noise and fireflies. I think it's much more efficient to render 2 (or more) passes required resolution (for example 640x480) and merge them together usind mode when less bright pixel is chosen (it's called "Darken" in Photoshop) than manual FSAA (Full Screen Anti Aliasing) method. In first case 614.400 pixels is rendered (640х480х2passes), in second case 1.228.800 (1280x960 - double resolution). Similar method is used in maxwell distributive render - each render node computes same frame (but with different seed) then all pictures are merged in one frame (in this case arithmetic mean based method is used I think). "Darken" method was already discussed here but it's really annoying to merge images manually - could it be done automatically?
I attached some pictures
We can see what in this case FSAA 2x isn't enough - image still too noisy.
Re: Multipass feature
Posted: Fri Aug 20, 2010 6:04 pm
by Chris
This could indeed be a solution for faster "de-noising". Nice idea

Maxwell already does this automatically with dist.rendering so i it should be possible to do with Octane also. One obstacle would be to have several framebuffers in vram at once.
MLT will probably solve the fireflies issue.
Chris
Re: Multipass feature
Posted: Fri Aug 20, 2010 6:15 pm
by jbeau3d
very interesting and nice pig. I'm going to have to implement this in my workflow until a proper solution can be figured out. Thanks for sharing:)
Re: Multipass feature
Posted: Sat Aug 21, 2010 6:58 am
by sam75
I thought about something like this too
would it be possible for octane to calculate twice the scene at the same time ? I know it would be slower but would it take twice the amount of memory ? if not i think it could be a good option
Re: Multipass feature
Posted: Sat Aug 21, 2010 9:11 am
by Chris
sam75 wrote:I thought about something like this too
would it be possible for octane to calculate twice the scene at the same time ? I know it would be slower but would it take twice the amount of memory ? if not i think it could be a good option
It would take twice the frambuffer if i understand it correctly. Wich means twice as much vram.
Re: Multipass feature
Posted: Sat Aug 21, 2010 9:16 am
by sam75
Chris wrote:sam75 wrote:I thought about something like this too
would it be possible for octane to calculate twice the scene at the same time ? I know it would be slower but would it take twice the amount of memory ? if not i think it could be a good option
It would take twice the frambuffer if i understand it correctly. Wich means twice as much vram.
rman1974 option would be enough, perhaps we would need to be able to stop the rendering with a time limit and not only number of samples as it is for now.
Re: Multipass feature
Posted: Sat Aug 21, 2010 10:44 am
by Chris
Yes a time limit would be nice to have.
Chris
Re: Multipass feature
Posted: Sat Aug 21, 2010 1:58 pm
by rman1974
Chris wrote:Yes a time limit would be nice to have.
Chris
I'm dreaming about time limit option in octane because it's much more handy for me than sample limit. Most of unbiased renders have time limit - I don't know why octane doesn't. Of course, time limit has a dark side - once I was rendering animation via network and I set time limit instead sample limit. At the end I got flicker in animation because render nodes isn't equal in cpu power and they give different results in the same time.
Re: Multipass feature
Posted: Sat Aug 21, 2010 7:35 pm
by sam75
rman1974 wrote:Chris wrote:Yes a time limit would be nice to have.
Chris
I'm dreaming about time limit option in octane because it's much more handy for me than sample limit. Most of unbiased renders have time limit - I don't know why octane doesn't. Of course, time limit has a dark side - once I was rendering animation via network and I set time limit instead sample limit. At the end I got flicker in animation because render nodes isn't equal in cpu power and they give different results in the same time.
for this we would need octane to save the solution on HDD and be able to resume it if needed, something like mxi files in maxwell.