That's a cool idea, importance mask. In theory, the PMC kernel is supposed to analyze the samples and determine which areas need more samples than others. But that probably takes longer calculate than a mask that says "just put more samples here." In MODOs preview render, you can hover your cursor over where you want more samples and it will give priority to that area while it's rendering. I got so used to using this feature that I find myself trying to do that in the Octane viewport and briefly wondering why nothing is changing.Stahlwolle wrote:Agree, but i wish there were a feature like "region-Painting" where you could after renderstop paint over some areas and the render engine just renders those painted/marked pixels - like region render just with an paintable mask and the continue-render option. that would be great cause you could focus the render onto the hard to render spots manually and focus the power to the spots where its needed. that would be a manual bias
Optional Biasing
Forum rules
Please add your OS and Hardware Configuration in your signature, it makes it easier for us to help you analyze problems. Example: Win 7 64 | Geforce GTX680 | i7 3770 | 16GB
Please add your OS and Hardware Configuration in your signature, it makes it easier for us to help you analyze problems. Example: Win 7 64 | Geforce GTX680 | i7 3770 | 16GB
In v2.0 you can use the region render to help clean up noisy areas in a render, but obviously only within a rectangular selection.riggles wrote:That's a cool idea, importance mask. In theory, the PMC kernel is supposed to analyze the samples and determine which areas need more samples than others. But that probably takes longer calculate than a mask that says "just put more samples here." In MODOs preview render, you can hover your cursor over where you want more samples and it will give priority to that area while it's rendering. I got so used to using this feature that I find myself trying to do that in the Octane viewport and briefly wondering why nothing is changing.Stahlwolle wrote:Agree, but i wish there were a feature like "region-Painting" where you could after renderstop paint over some areas and the render engine just renders those painted/marked pixels - like region render just with an paintable mask and the continue-render option. that would be great cause you could focus the render onto the hard to render spots manually and focus the power to the spots where its needed. that would be a manual bias
T.
Win10 x64|i7-9750H 2.6 GHz|32 GB RAM | RTX2080 max Q 8GB
- MaTtY631990

- Posts: 754
- Joined: Thu Apr 08, 2010 8:38 pm
I know PMC is the slowest kernel but the parameters option maxrejects can be used to control the bias of the result and in return should cut the render time down. I have tried it myself , but to no avail.
Has anyone else had luck with this setting.
Has anyone else had luck with this setting.
I was thinking this exact same thing! Been using region render for that, but it's obviously a little tough when it has to be a rectangle.Stahlwolle wrote:Agree, but i wish there were a feature like "region-Painting" where you could after renderstop paint over some areas and the render engine just renders those painted/marked pixels - like region render just with an paintable mask and the continue-render option. that would be great cause you could focus the render onto the hard to render spots manually and focus the power to the spots where its needed.
FYI, this isn't what "bias" (http://en.wikipedia.org/wiki/Bias_of_an_estimator) means.Stahlwolle wrote: that would be a manual bias![]()
Core i7-3770 / 16GB RAM / 2x GTX 780 6GB / Windows 7
This would be great!Stahlwolle wrote:Agree, but i wish there were a feature like "region-Painting" where you could after renderstop paint over some areas and the render engine just renders those painted/marked pixels - like region render just with an paintable mask and the continue-render option. that would be great cause you could focus the render onto the hard to render spots manually and focus the power to the spots where its needed. that would be a manual bias
Check out the pictures gabrielefx posted here http://render.otoy.com/forum/viewtopic.php?f=27&t=40861
After ~ 50 minutes/12bounces/8000smp/1600x900px with 4 Titans there is still so much grain left;
that`s is truely sad!
So, some region-painting would help a lot with those kind of scenes.
i think iray wanted to introduce something like that; not sure if it ever came out...
Intel i7-970 @3,20 GHz / 24 GB RAM / 3 x EVGA GTX 580 - 3GB
- Stahlwolle

- Posts: 68
- Joined: Sat May 19, 2012 9:36 am
- Location: Cologne, Germany
hey guys,
after a few testings with the region-render tool i can tell that its doing pretty same thing as i wished with region-painting. okay, its no organic form but at least it does not restart the complete render so you can stop your render after most of the image is noisefree and focus the gpu on the hard-to render regions with the tool - thats a big step forward in my opinion cause you dont have to sample the clean regions any more - maybe an algorithm could do that job in further versions of octane? im no expert but maybe such an algorithm could check the noise (difference in brightness and colour) - chk if its the texture or if its render-grain and focus the render to the pixels where grain is spottet. i know octane is a very "young" programm and i think the teams doing fine with it - wonder more about the effectivity-loss of the cudas - sounds a bit like "sinking boat" if the 580 gtx is same speed as a titan...
after a few testings with the region-render tool i can tell that its doing pretty same thing as i wished with region-painting. okay, its no organic form but at least it does not restart the complete render so you can stop your render after most of the image is noisefree and focus the gpu on the hard-to render regions with the tool - thats a big step forward in my opinion cause you dont have to sample the clean regions any more - maybe an algorithm could do that job in further versions of octane? im no expert but maybe such an algorithm could check the noise (difference in brightness and colour) - chk if its the texture or if its render-grain and focus the render to the pixels where grain is spottet. i know octane is a very "young" programm and i think the teams doing fine with it - wonder more about the effectivity-loss of the cudas - sounds a bit like "sinking boat" if the 580 gtx is same speed as a titan...
i7 3930K 6x3.2GHz | 32GB | ASUS Rampage Extreme iv | Geforce Titan + Geforce 780 ti | 1200W PSU
- Stahlwolle

- Posts: 68
- Joined: Sat May 19, 2012 9:36 am
- Location: Cologne, Germany
pardon... 590
i7 3930K 6x3.2GHz | 32GB | ASUS Rampage Extreme iv | Geforce Titan + Geforce 780 ti | 1200W PSU
That's basiclly what the PMC kernel already does. It examines the sample values and focuses more samples in high-contrast areas while lowering sample priority in low contrast areas.Stahlwolle wrote:maybe an algorithm could do that job in further versions of octane? im no expert but maybe such an algorithm could check the noise (difference in brightness and colour) - chk if its the texture or if its render-grain and focus the render to the pixels where grain is spottet.
- Stahlwolle

- Posts: 68
- Joined: Sat May 19, 2012 9:36 am
- Location: Cologne, Germany
for sure? can someone from octane-renderteam can verify this? if this would be true, wheres the value to adjust the threshold?
i know the slider to adjust the focus on the primary and secondary bounces but i would wonder a lot if the pmc would differ beyond that - mean: it says sample per pixel and that says a lot - i think thats so bruteforce as bruteforce can be..
i know the slider to adjust the focus on the primary and secondary bounces but i would wonder a lot if the pmc would differ beyond that - mean: it says sample per pixel and that says a lot - i think thats so bruteforce as bruteforce can be..
i7 3930K 6x3.2GHz | 32GB | ASUS Rampage Extreme iv | Geforce Titan + Geforce 780 ti | 1200W PSU
Here's link to a white paper on PMC (Population Monte Carlo) path tracing. It's an adaptive algorithm, not a simple threshold that you control with a slider.
http://pages.cs.wisc.edu/~yu-chi/resear ... pmc-pt.pdf
http://pages.cs.wisc.edu/~yu-chi/resear ... pmc-pt.pdf