GPU Pro 7

1. Octree Mapping from a Depth Camera 267

1.3.4 Managing Memory

The sparse octree used to represent a reconstructed 3D map will quickly grow

too large to fit entirely in GPU memory. Reconstructing a typical office room

at 1 cm resolution will often take as much as 6–8 GB. Use of a GPU with more

memory will allow for larger scenes at higher resolutions, but there will always

be applications where a physical memory increase is not practical to meet the

requirements.

To handle this, we developed an out-of-core memory management framework

for the octree. At first glance, this framework is a standard stack-based octree on

the CPU. However, each node in the tree ...