How would you decide how many bounces is enough or when to stop because for now it just seems like you test a certain number of bounces until one of them produces the best output? Or am I just missing something?
At some point the additional term Kn(Le)K^n(Le)Kn(Le) will be negligible compared with all the previous terms. That is the time to stop. Of course, how "negligible" is negligible is totally up to you.
Is there a number of bounces that is generally used in practice? Maybe something like a changing value for resolution?
In lecture professor showed how we actually take samples that can be infinite in the number of bounces yet still be unbiased. This is because if we always use a limited number of bounces k, then we will always be underestimating the light in the scene.
I heard the professor respond to a question about depicting a room entirely made of mirrors, where the light bounces infinitely. In this case for the rendering equation we only consider bounces up to a certain point?