Questions about metroplis sampling.

Practical and theoretical implementation discussion.
shiqiu1105
Posts: 138
Joined: Sun May 27, 2012 4:42 pm

Re: Questions about metroplis sampling.

Post by shiqiu1105 » Fri Oct 03, 2014 6:13 pm

Also, about Veach's weighting trick.

In the warm up phase, n samples x1...xn are selected with probability p(x), and later we should weight all the contributions by
(f(x1)/p(x1) +...+f(xn)/p(xn)) / n right? I don't see people doing this when calculating the final contribution.. Is my understanding somehow wrong?

For better printed equations about this question, see this post: viewtopic.php?f=3&t=1992

friedlinguini
Posts: 89
Joined: Thu Apr 11, 2013 5:15 pm

Re: Questions about metroplis sampling.

Post by friedlinguini » Fri Oct 03, 2014 9:46 pm

MohamedSakr wrote:sorry for jumping over the thread, but I wonder how ERPT works?

my previous knowledge so far, how PT works, BDPT, VCM, PPM, PM + final gather, MLT (BDPT), MLT + MIS (MMLT)

how ERPT is different? from the name it seems to apply only to PT, but can it apply to BDPT or VCM like algorithms?
does it converge faster than MLT?

and typical limitations? "stuck on peaks, caustics failure, glossy reflection artifacts, etc..."

feel free to clarify as much as possible :D
Let me Google that for you. :-)

http://atlas.cse.ohio-state.edu/WWW/Sig ... -cline.pdf

The basic idea is to start with per-pixel path tracing, but instead of splatting a single sample with brightness proportional to the energy of the path, generate mutations of the path, with the number of mutations proportional to the original energy. Hence, redistributing the energy of the traced path. Because it tries to stratify across the image plane, I expect that it has a somewhat harder time finding caustics than MLT with explicit eye connections. The paper describes some bias-introducing noise-reduction techniques, but they seem to be equally applicable to ordinary MLT.

MohamedSakr
Posts: 83
Joined: Thu Apr 24, 2014 2:27 am

Re: Questions about metroplis sampling.

Post by MohamedSakr » Fri Oct 03, 2014 11:33 pm

friedlinguini wrote:
MohamedSakr wrote:sorry for jumping over the thread, but I wonder how ERPT works?

my previous knowledge so far, how PT works, BDPT, VCM, PPM, PM + final gather, MLT (BDPT), MLT + MIS (MMLT)

how ERPT is different? from the name it seems to apply only to PT, but can it apply to BDPT or VCM like algorithms?
does it converge faster than MLT?

and typical limitations? "stuck on peaks, caustics failure, glossy reflection artifacts, etc..."

feel free to clarify as much as possible :D
Let me Google that for you. :-)

http://atlas.cse.ohio-state.edu/WWW/Sig ... -cline.pdf

The basic idea is to start with per-pixel path tracing, but instead of splatting a single sample with brightness proportional to the energy of the path, generate mutations of the path, with the number of mutations proportional to the original energy. Hence, redistributing the energy of the traced path. Because it tries to stratify across the image plane, I expect that it has a somewhat harder time finding caustics than MLT with explicit eye connections. The paper describes some bias-introducing noise-reduction techniques, but they seem to be equally applicable to ordinary MLT.
thanks a lot :) , I really searched for the paper, but didn't find the link!!

Post Reply