First attempt at Kelemen style Metropolis

Practical and theoretical implementation discussion.
Post Reply
dylan
Posts: 7
Joined: Sun Sep 23, 2012 7:16 pm
Location: Cambridge, UK

First attempt at Kelemen style Metropolis

Post by dylan » Sun Sep 23, 2012 7:39 pm

As well as the title, this is my first attempt at a path tracer.
My standard sampler code performs quite nicely, but my Metropolis sampler, which is copied nearly verbatim from Csaba's paper, is definitely wrong. It looks like old samples are contributing at new sample locations, or vice versa, and edges seem to be contrast enhanced, even wrapping around borders (note the bottom edge of the first image). Most of my test images seem to get darker (below the true value) as the sample count increases.

My not-quite-Metropolis sampler:
http://i.imgur.com/hbl8p.jpg
Image
My path traced, noisy reference:
http://i.imgur.com/7vlzQ.jpg
Image

Any ideas? I still can't see how my code is managing to clearly put samples at the wrong locations.

dylan
Posts: 7
Joined: Sun Sep 23, 2012 7:16 pm
Location: Cambridge, UK

Re: First attempt at Kelemen style Metropolis

Post by dylan » Thu Oct 04, 2012 1:37 pm

The "stars" were a result of an uninitialised variable (well that's embarrassing). The contrast enhancement was due to a weighting problem, solved in this thread.
Here's an updated fixed render:
Image

Does that refraction look right? I'm using an IOR of 1.5.

Post Reply