Occlusion Culling

[Halifax]

Really, I don't agree at all with occlusion culling. Although it is easy to implement for beginners, etc. I don't think of it as a good solution. The good solution right now is PVS algorithms, in particular Insomniac Games algorithm. Although it is time intensive, 1 day to generate a PVS data set for a Resistance: FOM map. It is horribly efficient especially with their compression algorithm, which they say is great because the PVS set can be kept in the small memory of the PS3 much easier.

Basically, it's like a new BSP, but much more easier, and Al Hastings style. (Just a thought)

EDIT:
Alright extrusion was not nearly as expensive as I had previously thought, but it is still expensive. It would take 3 multiplications, and 4 additions per vertex. Which means 12 multiplications, and 16 additions per box extrusion, and you would just have to use the faceDir of the camera as the plane normal. The only possible problems with the cull boxes is still the O(n^2) (I think) checking of cull boxes.

[Dorth]

Halifax, get your hand on Game Programming Gem 5 and fast!
Not only is it a great book, but more than one gem (I think 3?) tackle this problem from a side or another. Basically, using their way (2 of which you can combine for sure, third I dunno) it'll go so mind-boggingly fast with occlusion that of course you'd like PVS, but it'll be an alternative, no longer an underdog

[Halifax]

Oh, by no means do I consider occlusion culling an underdog, I just don't really like it, although I have to say it is great for dynamic geometry which is the current downfall to PVS.