Implementing Vertex Connection and Merging (tech report)

[ingenious]

Hi all,

I have published a technical report that discusses the implementation of our vertex connection and merging algorithm:

Implementing Vertex Connection and Merging
http://www.iliyan.com/publications/ImplementingVCM

The focus is on the efficient path MIS evaluation. Special cases are discussed, such as correctly handling infinite lights and orthographic cameras, as well as point lights and specular materials. There's also a discussion on how to apply the weight scheme to bidirectional photon mapping and bidirectional path tracing. In the latter case, the scheme reduces to something almost equivalent to Dietger's. The slight differences are because the scheme was independently derived (but I obviously couldn't publish it before the VCM paper).

SmallVCM has also seen an update with more comments as well as references to the specific equations in the tech report, with the hope to make both the implementation and the tech report easier to parse. There have also been discussions here lately on the eye connection technique and the associated pdfs, so interested people might want to check out this part of code, which has been refactored and more heavily commented.

[graphicsMan]

Thanks Iliyan. I look forward to reading over this.

[Dade]

SmallVCM

I simply love SmallVCM, it is the renderer with the easier to read sources I have ever seen.

[spectral]

Congratulation, and thanks for sharing and improving the details and documentation

[ingenious]

Hi guys,

I recently updated the VCM tech report to revision 2. This important release includes some bug fixes in the formulas, and new section discussing the use of per-pixel merging radii as well as a memory-efficient variant of progressive VCM which reduces the memory footprint to that of PPM.

Also, if you're attending SIGGRAPH this year, make sure not to miss our course "Recent Advanced in Light Transport Simulation: Theory and Practice". There's lots of interesting stuff in the syllabus: Jaroslav will cover the path-space Monte Carlo, and I will talk about VCM, including some more best VCM practices that I've collected in the past year. Anton will then dive deep into Markov chain Monte Carlo, and Juan will bring us back to the reality of integrating pretentious research awesomeness into a production renderer. See you there!

[spectral]

Thanks a lot Ingenious,

I will take a look at revision 2 asap, but I have 2 questions :
1) Which formulas have been fixed ?
2) Have you also fixed SmallVCM ?

Once read, I will take a look at your Sylabus... good job

Thanks

[ingenious]

1) Which formulas have been fixed ?
2) Have you also fixed SmallVCM ?

1) A few equations, mostly in Section 4.2. I can't recall which ones at the moment, but it may be a good idea to add a history of changes as an appendix in the end.

2) It was only the few formulas in the tech report that had bugs. The implementation in SmallVCM code has been correct for some time now - it's been tested much more than the math

[ypoissant]

Thanks ingenious,

Since I can't attend SIGGRAPH, I hope you can post the course somewhere after SIGGRAPH.

[ingenious]

Sure thing, slides will be available online soon.

[ypoissant]

Have the slides or course notes been posted somewhere yet?