RealmDreamer: Text-Driven 3D Scene Generation (realmdreamer.github.io)

I don't really get why they need the occlusion volume? What is its functionality? Doesn't rendering a pointcloud from a novel viewpoint already give the (dis-)occlusion mask?