I recently read through the free RTGII chapter about reservoir sampling. I’ve also noticed that resampled importance sampling has been gaining popularity (e.g. ReSTIR). These both seem like useful techniques when sampling direct lighting, but it seems difficult to re-compute PDFs for the sake of MIS when a light has been sampled through a BSDF sample.
For example, in the case of resampled importance sampling, we don’t know what the other samples in the list would be. Could you provide any insights on this problem?
Interesting question indeed - had to think about that for a while, and
actually had to write two different answers :-).
One possible answer is that sample importance resampling is only generating samples relative to an existing PDF (that you choose to
take these samples), so it should be possible to evaluate that
existing PDF to compute PDF values for other samples (in an MIS context).
I’d have to think a bit more about it, though, as it’s a tricky
technical question indeed. I’d be happy to chat more about that offline
not sure if the forum is the right format for this question… feel
free to contact me offline any time.