Comments on: Multiple Importance

By: Joe Mwangi

Joe Mwangi — Mon, 19 Jul 2010 09:55:01 +0000

To be frank with you I was not really conversant with MIS in BDPT. After really reading it http://www.maw.dk/3d_graphics_projects/downloads/Bidirectional%20Path%20Tracing%202009.pdf together with veach thesis, I think I have a clear understanding of the importance of using MIS. I think RIS won’t be efficient in this area since it requires many pdf evaluation i.e large number of path length. To make it worthwhile to use it. I see no reason of using RIS. “What would I be resampling?” – The pdf’s of the path you could have likely generated, and reject the pdfs that give bad contribution.

By: Simon Brown

Simon Brown — Wed, 09 Jun 2010 11:58:13 +0000

Not tried. I’m not actually so familiar with that paper, but it seems to be related to drawing samples for BRDFs or light sources that have no well-matched distribution to use directly for IS. Since all IS in that scene is trivial (hemi-spherical emission, and Lambertian BRDF), what would I be resampling?

By: Joe Mwangi

Joe Mwangi — Wed, 09 Jun 2010 08:51:55 +0000

Have you tried Resampled Importance Sampling by David Cline & Justin Talbot. Coupled with MIS it reduces the image variance by margins with half the required samples/pixel.

By: Anders Langlands

Anders Langlands — Fri, 06 Feb 2009 10:04:55 +0000

Great blog! There was some discussion about LDR vs HDR anti-aliasing over at ompf a while ago. Coming from a visual effects background, my instinct is always to preserve the dynamic range as long as possible, which means anti-aliasing in HDR.

I’ve never really been able to think of a way of getting a good result when looking directly at a lightsource. Of course, in reality, you would never look directly at a lightsource to begin with – this would result in glare and lens flares. The cornell box doesn’t have a 2D plane area light sitting at the top of it, so I think the aliasing problem is likely one of our own construction

By: Simon Brown

Simon Brown — Fri, 06 Feb 2009 08:27:38 +0000

As a technique for rendering Cornell boxes, path tracing with explicit light sampling is not bad. It just so happens that the one type of path this technique generates is common in the Cornell box scene, so doesn’t have crazy variance.

However, as you suggest, as soon as you start introducing different BRDFs, or even just strong indirect lighting, this one path type isn’t as strong as other path types, so you start to get much better results by generating paths bidirectionally.

By: Kevin Beason

Kevin Beason — Fri, 06 Feb 2009 05:33:08 +0000

Just discovered your blog.

Results look good! Seems like you would get similar results from just using path tracing with explicit light sampling. However if there was a specular surface somewhere then this would really help.

I hope to implement this some day. Your results are encouraging Also, a very clear description. Thank you.