They create occluder meshes to partition a large NeRF into multiple smaller ones. That way, stepwise traversing through empty space in the NeRF can be avoided.

They create occluder meshes to partition a large NeRF into multiple smaller ones. That way, stepwise traversing through empty space in the NeRF can be avoided. But I'd guess that DepthNeRF would still be more efficient overall, because their partitioning might actually increase overdraw unless they reach an opaque alpha and the UE occlusion engine works pixel perfect.