Lecture 13: Global Illumination & Path Tracing (44)
CeHao1
An interesting topic is how many bounces should we take for rendering. A good criteria is that if the light intensity decayed to a very small amount.
But in the next glass ball example, it seems even after many bounces, the light is still very visible. So can we use some way to approximate the residue of all higher-order terms? Maybe evaluating the maximum of it so that we can stop the bouncing somewhere.
Rishiparikh
I agree with this, I think that intuitively, objects with different reflectivities might need more or less number of bounces to closely represent how it might look in the real world. I'm also curious if the overall background brightness affects when we should stop evaluating bounces. If we have a very dark room, dim amounts of light will make a big difference.
StephenYangjz
It's interesting to me that the summation works out intuitively. We previously defined L to be the emitted light + the reflected light at this specific angle of the hemisphere -- which would be recursive in its definition. The Neumann series nicely expands it and allows us to see that the equation equals the sum of all the N bounces of reflections and also allows for future possibilities of Monte Carlo Integration.
An interesting topic is how many bounces should we take for rendering. A good criteria is that if the light intensity decayed to a very small amount.
But in the next glass ball example, it seems even after many bounces, the light is still very visible. So can we use some way to approximate the residue of all higher-order terms? Maybe evaluating the maximum of it so that we can stop the bouncing somewhere.
I agree with this, I think that intuitively, objects with different reflectivities might need more or less number of bounces to closely represent how it might look in the real world. I'm also curious if the overall background brightness affects when we should stop evaluating bounces. If we have a very dark room, dim amounts of light will make a big difference.
It's interesting to me that the summation works out intuitively. We previously defined L to be the emitted light + the reflected light at this specific angle of the hemisphere -- which would be recursive in its definition. The Neumann series nicely expands it and allows us to see that the equation equals the sum of all the N bounces of reflections and also allows for future possibilities of Monte Carlo Integration.