Metric for selective voxel activation within VDB bricks

Develop a perceptual or similarity-based metric for selecting which voxels to activate within each 32×32×32 OpenVDB leaf brick during fixed-rate compression of structured-regular volumes, so that partial brick activation can control block artifacts while meeting a specified compression rate.

Background

The paper proposes a fixed-rate compression algorithm that partitions structured-regular volumes into 32×32×32 bricks (matching OpenVDB leaf nodes), sorts bricks by similarity to a background value, and activates all voxels in a subset of bricks to meet a target compression rate. Activating all or none of a brick’s voxels can produce visible block artifacts.

The authors suggest an alternative of partially activating voxels within each brick based on similarity, but this requires a metric to decide which voxels to activate while maintaining the fixed-rate target. They did not implement such a metric and explicitly leave it as future work.

References

Another option would be to decide the number of voxels we want to activate from the brick based on similarity, too, e.g. that starting at a certain threshold we activate half the number of voxels but twice the number of bricks. We leave such optimizations as future work as we would require a (possibly perceptual) metric telling us which voxels to activate.

— GPU Volume Rendering with Hierarchical Compression Using VDB  (2504.04564 - Zellmann et al., 6 Apr 2025) in Section 3.2.1 (Compression)