RR Probability and Image Noise
Posted: Tue Sep 02, 2014 7:40 am
Hi,
I came across something that confuses me a bit.
I used PMC for water reflections of nearly 1oo% and got bad and splotchy results for the reflection. In fact, the reflection hardly resembled the object to be reflected.
After testing each PMC parameter, it turned out that RR Probability has the most influence on how well reflections are rendered. Caustics blur as well, but it just blurres it and the reflection still does not resemble the object.
I am familar with the concept of russian roulette in path tracing. However, in Octane, the closer I set RR to 1, the better the reflections, i.e. the most uniform the noise. With low values, or left to 0, the reflection get very splotchy, but with high values, the reflection almost accumulates like the noise in the 'to-be-reflected' object itself. In my understanding, high RR Probability (close to 1) means it is very likely that paths are randomly killed, say with russian roulette 0.5 there is a 50% chance that the ray gets randomly killed.
Therefore I expected: higher RR Values -> more random path kills -> less rays finish to the end -> more splotchy reflection because rays are missing -> and also shorter render time.
In Octane it is exactly the opposite. Does RR probability in Octane actually mean a higher probabability that rays do not get randomly killed? So a value close to 1 would mean almost all rays finish their bounces.
Is it also like a well know thing in Octane, that RR at high values is required for less noise in PMC?
edit: got it
I came across something that confuses me a bit.
I used PMC for water reflections of nearly 1oo% and got bad and splotchy results for the reflection. In fact, the reflection hardly resembled the object to be reflected.
After testing each PMC parameter, it turned out that RR Probability has the most influence on how well reflections are rendered. Caustics blur as well, but it just blurres it and the reflection still does not resemble the object.
I am familar with the concept of russian roulette in path tracing. However, in Octane, the closer I set RR to 1, the better the reflections, i.e. the most uniform the noise. With low values, or left to 0, the reflection get very splotchy, but with high values, the reflection almost accumulates like the noise in the 'to-be-reflected' object itself. In my understanding, high RR Probability (close to 1) means it is very likely that paths are randomly killed, say with russian roulette 0.5 there is a 50% chance that the ray gets randomly killed.
Therefore I expected: higher RR Values -> more random path kills -> less rays finish to the end -> more splotchy reflection because rays are missing -> and also shorter render time.
In Octane it is exactly the opposite. Does RR probability in Octane actually mean a higher probabability that rays do not get randomly killed? So a value close to 1 would mean almost all rays finish their bounces.
Is it also like a well know thing in Octane, that RR at high values is required for less noise in PMC?
edit: got it