Lecture 13: Global Illumination & Path Tracing (27)
mylinhvu11
For clarification, what's the difference between sampling the BRDF versus the Lights? The pseudocode of the two functions seems very similar except for the implementation of solving for L which leads me to believe these functions are the same except possibly scaled differently due to the L variable.
ncastaneda02
I think the main difference is that when you are using the BRDF, you have to explicitly calculate your radiance separately from getting your sample direction due to how the sample function works. When working with lights directly, you get the radiance for "free" when calculating your sample direction. These radiance values will likely end up being slightly different, but I think that they could end up corresponding pretty well with one another depending on your implementation.
jacklishufan
In practice how do we actually sample points on the light source for area light? One disadvantage of of uniform hemisphere sample is that most of the "work" do not have significant contribution in the final output. Imagine the case when we have a glowing ball as light source with a color map over the L, then it only makes sense to sample the points on the light surface that is facing towards the object, not opposite, or we have 1/2 chance of getting 0.
Ishaandham19
One example of when importance sampling of lights vs importance sampling of BRDF could yield varying results is in the case of a
highly polished surface for which sampling BRDF would yield better results than sampling light sources since the reflections would be better captured.
For clarification, what's the difference between sampling the BRDF versus the Lights? The pseudocode of the two functions seems very similar except for the implementation of solving for L which leads me to believe these functions are the same except possibly scaled differently due to the L variable.
I think the main difference is that when you are using the BRDF, you have to explicitly calculate your radiance separately from getting your sample direction due to how the sample function works. When working with lights directly, you get the radiance for "free" when calculating your sample direction. These radiance values will likely end up being slightly different, but I think that they could end up corresponding pretty well with one another depending on your implementation.
In practice how do we actually sample points on the light source for area light? One disadvantage of of uniform hemisphere sample is that most of the "work" do not have significant contribution in the final output. Imagine the case when we have a glowing ball as light source with a color map over the L, then it only makes sense to sample the points on the light surface that is facing towards the object, not opposite, or we have 1/2 chance of getting 0.
One example of when importance sampling of lights vs importance sampling of BRDF could yield varying results is in the case of a highly polished surface for which sampling BRDF would yield better results than sampling light sources since the reflections would be better captured.