Update - 2014-07-17
I will be giving a talk at Siggraph this year, Implementing Efficient Virtual Shadow Maps for Many Lights,
where I will try to give a more practical approach to the techniques presented in the paper. I intend to highlight important engineering aspects that led to good performance, and present details
that show how the implementation works.
Update - Slides
I have now published my slides from the presentation of the paper at I3D, they contain a lot of animations and since recompression comes out terrible in powerpoint i have prepared a
pdf version, which is a lot smaller. I have not really had time to go over the notes, but I hope they are relevant.
At last, I got around to have a go at the problem of supporting shadows with many lights. It came about by me starting out pondering all manner of new and exciting ways to compute shadows,
at the beginning really considering shadow maps impractical for many-light applications. This, in a way, appeared to be the recieved wisdom in the field. However, as I thought about
and discussed the problem with my colleagues, I realized I could not substantiate such a rejection. At this point I began to seriously consider how to make shadow maps work,
work efficiently, and produce high-quality results. This is the resulting paper, and while my answer has many ingredients, I hope it is understandable and compelling, and ultimately useful.
The paper is not complete without the Supplementary Video, which shows the results and camera paths used in the performance measurements in the paper.
I'd really like to offer promises about demo code, but my time as a PhD student is nearly up and the future uncertain, so it could be some time before I get around to it.
And yes, there are multiples of the Tardis flying around in necropolis. I had to have something dynamic casting shadows!