Previously, when pre-existing meshes were provided to the simplifier, they were in physical coordinates and therefore often 30-40x more spaced out than the voxel versions. For very large meshes, this can push them past the limits the packed coordinate representation can tolerate (2^20 for uint64_t). This caused gross corruption in large simplified meshes, the obvious use case.
Now we divide by the voxel resolution before simplifying, which increases the range considerably
Previously, when pre-existing meshes were provided to the simplifier, they were in physical coordinates and therefore often 30-40x more spaced out than the voxel versions. For very large meshes, this can push them past the limits the packed coordinate representation can tolerate (2^20 for uint64_t). This caused gross corruption in large simplified meshes, the obvious use case.
Now we divide by the voxel resolution before simplifying, which increases the range considerably
.