Just a quick post.
In my last post I mentioned
Deferred Radiance Transfer Volumes and in that presentation they mentioned that they blur the volume in order to soften the irradiance and my first thought was, why on earth didn't I think of that? So I tried the blur (while in the midst my computer kept crashing because it was overheating; I really need to upgrade my machine) and these are the results:
 |
| Unblurred |
 |
| Blurred |
As you can see, the irradiance aliasing has completely disappeared! Also, some of the sampling artifacts I was having before turned out to be caused by a bug in my voxelization process. I was rendering the scene in 6 directions and counting backfaced pixels and using a threshold to determine if the voxel is inside or outside the geometry (I got the idea from the
Robust Inside and Outside Solid Voxelization post on Nick Darnel's blog), but it turned out I was accidentally using the number of cells in the irradiance volume instead of the number of pixels in my sample as a normalization factor. Oops! Well, this definitely explained why the number of artifacts increased as I increased the resolution of the irradiance volume :)
