ompf2.com

I read some articles on many light rendering (or instant GI/radiosity).

I find it hard to understand its efficiency. It seems to me like using length one camera path connecting to many light sub-path with arbitrary length and an average weighting term. Isn't is better if we use MC/QMC pre-generated BDPT light sub-path with basic MIS?

Any good reason to use VPLs rather than BDPT?

Used naively, probably not, but there are advantages to amortizing the cost of light paths over many eye paths. IGI maps well to GPUs--you path-trace light paths, but can use rasterization and shadow maps instead of eye paths and shadow rays. If you have a lot of VPLs, you can prune the set judiciously (e.g., lightcuts, row-column sampling, etc.) to get the advantages of having a whole bunch of lights without actually having to calculate the contributions for all of them, whereas BDPT gives you one light path per eye path that might or might not provide a significant contribution.

Things start getting messy when you figure in glossy/specular BSDFs, but then again I don't think a universally "good" (e.g., consistent, useful for both previews and final rendering, memory-bounded, handles arbitrary massive scenes with arbitrary BSDFs, etc.)