Sobol' sequences

Practical and theoretical implementation discussion.
andersll
Posts: 22
Joined: Thu Mar 21, 2013 4:34 pm

Re: Sobol' sequences

Post by andersll » Mon Sep 28, 2015 11:13 pm

Yeah I think CMJ performs pretty well from what I've seen, but still suffers from the limitation that you've got to know your sample count up front.

cessen
Posts: 34
Joined: Sun Apr 14, 2013 9:34 pm

Re: Sobol' sequences

Post by cessen » Tue Sep 29, 2015 5:51 am

Okay, I have some more sample test images. Here is (14,15) at 64, 256, and 65536 samples, respectively:

Image
Image
Image

And here is (24,25):
Image
Image
Image

So, both of them perform reasonably well across a wide range a sampling rates, it seems. Though neither are great at 64.

However, I also checked some other dimension combinations, and found that (14,24) performs poorly:
Image
Image
Image

So it has its problems, too. But even so, I find the lower dimensions to be better distributed with Faure-permuted Halton than with Sobol. Though, admittedly, I haven't done a super thorough investigation. ;-)

Post Reply