Bug #344

rough* materials broken with veach style MLT

Added by Adam Celarek over 2 years ago. Updated over 2 years ago.

Status:FeedbackStart date:11/13/2015
Priority:NormalDue date:
Assignee:-% Done:

0%

Category:-
Target version:-

Description

Hi,
rough* materials produce a lot of noise with veach's mutations in MLT. They are unusable in the current state. I'm unsure whether this is due to the algorithm or due to some bug in the code.

See the attached files, the changes are limited to replacing rough* materials with almost equivalent smooth materials (roughconductor -> conductor, roughcoating -> coating etc).

working.jpg (488 KB) Adam Celarek, 11/13/2015 02:51 pm

broken.jpg (508 KB) Adam Celarek, 11/13/2015 02:51 pm

veachDoor.zip (1.77 MB) Adam Celarek, 11/13/2015 02:51 pm

History

#1 Updated by Adam Celarek over 2 years ago

the attached images were rendered with 250 samples per pixel. the rough* materials version didn't even converge after over 7 million samples per pixel.

#2 Updated by Wenzel Jakob over 2 years ago

  • Status changed from New to Feedback

Hi Adam,

it's not a bug -- that's just how this method works. Veach-MLT lumps any non-prefectly-specular materials into the same category, and the mutations essentially treat them using the strategies developed for diffuse-ish materials. This worked well for the simple scenes Veach was rendering in his thesis (mostly perf. specular with diffuse surfaces) but obviously it will give very poor results for the kinds of scenes that users are currently interested in with lots of glossy interreflection. The goal in Mitsuba was to reproduce the original method as closely as possible for comparisons, so this is intentional.

Wenzel

#3 Updated by Adam Celarek over 2 years ago

a bit surprising..
but ok, you can close :)

Also available in: Atom PDF