Archive for April, 2012
Bidirectional Instant Radiosity is the title of a paper by B. Segovia et al which presented a new sampling strategy to find virtual point lights (VPLs) that are relevant to the camera. The algorithm given for generating VPLs is:
- Generate \(N/2\) “standard” VPLs by sampling light paths (i.e. vanilla instant radiosity)
- Generate \(N/2\) “reverse” VPLs by sampling eye paths (compute their radiance using the N/2 standard VPLs)
These \(N\) VPLs are then resampled into \(N^\prime\) VPLs by considering their estimated contribution to the camera. Finally the \(N^\prime\) resampled VPLs are used to render an image of the scene.
In this post I’ll describe how I think this approach can be generalised to generate VPLs using all the path construction techniques of a bidirectional path tracer. As usual I’m going to assume the reader is familiar with bidirectional path tracing in the Veach framework.
I should state that this is an unfinished investigation into VPL sampling. I’m going to describe the core idea and formally define the VPL “virtual sensor”, but proper analysis of the results will be part of a future post (and may well indicate that this approach is not advisable).