All-in-one Integration

Merge all sources into a voxel block and re-render a depth map cloud for each source in a single kernel. This allows all voxel block storage to be removed, resolves a colour race condition, but will require a DIBR step for virtual view rendering along with a depth splat step. It also prevents inter-block checks unless the voxel blocks overlap by one voxel.

Instead of rendering depth maps directly, some storage could be used in the form of a bit mask indicating if a voxel is inside or outside a surface. This would involve 512 bits per occupied voxel block which may be less data than sending all the depth maps to remote machines. Further, surface colour information could some how be merged into a voxel block so that original RGB images don't need to be sent also.

Another disadvantage would be that SDF between frames is now lost so some alternative (MLS?) method of smoothing would be required. Past frames may still be used in some capacity, or an SDF might be regenerated.