Eric Lengyel told me that PVS (Potential Visibility Sets) is an outdated
technology for today’s high-level polygon count worlds. As far as I know,
PVS is simply precompiled visibility obtained from portals. Portals are the
empty space in between convex polyhedra, and portal rendering, as far as I
gather, are simply N-gon planar polygons extruded recursively from one
portal opening to another, clipped from the outer extends as it intersects
any polygon reaching those nodes during the extrusion.
Therefore, how can Portals, which are run-time, be more effective then
something precomputed? My only guess is that the data structure would be too
large to be feasible. Sort of if you have a very large lookup table,
sometimes the risks (say from virtual page faults to cache misses) would
override the benefits. But I am not sure if that is what he meant. Can
anyone point me in a direction for what modern high polygon 3D engines use
for scene occlusion / culling and why PVS trees are considered outdated - or
any resources that explain this specific concept in more detail?
Thanks,
Matthew