Is the tentative contribution a rough estimate before the ray actually goes out pathtracing and finds out about the actual incident luminance?
JimL
@outousan I don't think it's a estimation. In my understanding, we can assume the light is always hitting the surface. By this assumption, we do not need to ray trace back to the light source (finding if the ray is hitting anything). And if the contribution to the surface is already small, we can somehow discard it using the RR method to ensure the estimator is not unbiased. Not 100 percent sure if I am understanding this right.
allai5
How do you determine whether or not the sample is low-contribution enough to consider discarding?
KrystalTea
Just curious about why it is called 'Russian Roulette'? Is there any similarity?
yee
@allai5 I think it's a threshold decided by yourself. As the following slides suggests, we regard contributions with illuminance less than a certain threshold as low-contribution, and terminate xx% of those contributions. Increasing such threshold always leads to lower cost and worse render fidelity.
Lavender
@KrystalTea I think it's called Russian Roulette because we're gambling on that randomly discarding low-contribution samples would still have good rendering result
Is the tentative contribution a rough estimate before the ray actually goes out pathtracing and finds out about the actual incident luminance?
@outousan I don't think it's a estimation. In my understanding, we can assume the light is always hitting the surface. By this assumption, we do not need to ray trace back to the light source (finding if the ray is hitting anything). And if the contribution to the surface is already small, we can somehow discard it using the RR method to ensure the estimator is not unbiased. Not 100 percent sure if I am understanding this right.
How do you determine whether or not the sample is low-contribution enough to consider discarding?
Just curious about why it is called 'Russian Roulette'? Is there any similarity?
@allai5 I think it's a threshold decided by yourself. As the following slides suggests, we regard contributions with illuminance less than a certain threshold as low-contribution, and terminate xx% of those contributions. Increasing such threshold always leads to lower cost and worse render fidelity.
@KrystalTea I think it's called Russian Roulette because we're gambling on that randomly discarding low-contribution samples would still have good rendering result