So if I have dynamic voxel terrain data stored in an octree representation what would be the best way to go about converting the octree to a polygon mesh which I can render with OpenGL?
A lot of the meshing algorithms I have found online assume that voxel data is stored in a 3D Array however for the level of detail I am using storing chunk data in a massive array isn't really an option (chunk size 64x1024x64 with each block roughly 4bytes) split into several 64x64x64 small octrees, I could make the chunk size smaller but then that means more draw calls etc.
Basic searches on google for 'meshing an octree' seem to bring up lots of research on doing the inverse (converting from meshes to SVOs).
The problem with using these algorithms on the octree instead of the array is that the octree has a much slower access time not to mention the time cost of copying the data from the octree representation.
I have considered the possibility of converting to an RLE or spatial hashing system but these two systems both have their drawbacks especially since I plan to use the same octree when I come to implement collision detection.
I would guess that it would be best to traverse down the tree rather than iterating over it like an array, but some more ideas about how I could optimise this would be appreciated.