Practical and theoretical implementation discussion.
Post Reply
Posts: 42
Joined: Mon Jul 23, 2012 11:05 am


Post by Zelcious » Thu Sep 13, 2012 4:15 pm

Is it the mutated path that goes back into the re-sampling step or the original. If it's the mutation then is pmc-erpt unbiased? Should violate detailed balance right, since the number of mutation is dependent of path intensity.

Trivia: I did a little calculation last where it tried to come up with the simplest case where basing the mutation length on intensity violates detailed balance.
Two states with intensity 1,2 respectively does the trick if you let mutation length be equal to intensity.

Btw, are there any known methods dealing with the more relaxed general balance?

Posts: 50
Joined: Tue Nov 29, 2011 10:33 am


Post by Dietger » Mon Oct 08, 2012 3:01 pm

The paper couldn't convince me that PMC-ERPT is unbiased either (or even consistent for that matter). I recently reread the PMC sources referenced in the paper and I still don't see how the authors got from there to the presented PMC-ERPT algorithm. Aside from the lack of proofs, the authors skip over some details which makes it impossible for me to figure out how their method is supposed to work. Interesting observation: the papers Cornell box comparison of ERPT and PMC-ERPT seems to converge to different images...

PS. I initially tried to write down in more detail why the paper does not make sense to me, but I got too frustrated and gave up :) If anyone really wants to know I could give it another go. Or if someone can explain to me why I am an idiot and the paper makes perfect sense, please do!

Post Reply