After upgrading Nellie from v0.4.1 to v1, the voxel_matches.npy output appears to have changed in a way that breaks a downstream remapping workflow.
voxel_matches.npy is being used to map concentration fields from one frame to the next. In v1, many voxels from the current frame do not appear in the match array for that time point, and many repeated entries are present instead.
Observed pattern:
• voxels_next: N x 3
• voxels_prev: N x 3
• unique(voxels_next): M x 3
• unique(voxels_prev): K x 3
It is expected that M < N by a small constant factor, since duplicate mappings from previous to next voxels can occur. In v0.4.1, K ~ M was typically observed. In v1, K ~ M/3 is being observed instead, suggesting that many source voxels are missing from the mapping.
Impact
In simulation workflows, this causes some concentration fields not to be remapped to the next frame, leading to signal loss over time.
Additional notes
• No obvious issues were noticed in the reassigned TIFF that tracks from the beginning to the end of the movie.
• This suggests the problem may be specific to the voxel_matches.npy output rather than the broader tracking result.
After upgrading Nellie from v0.4.1 to v1, the voxel_matches.npy output appears to have changed in a way that breaks a downstream remapping workflow.
voxel_matches.npy is being used to map concentration fields from one frame to the next. In v1, many voxels from the current frame do not appear in the match array for that time point, and many repeated entries are present instead.
Observed pattern:
• voxels_next: N x 3
• voxels_prev: N x 3
• unique(voxels_next): M x 3
• unique(voxels_prev): K x 3
It is expected that M < N by a small constant factor, since duplicate mappings from previous to next voxels can occur. In v0.4.1, K ~ M was typically observed. In v1, K ~ M/3 is being observed instead, suggesting that many source voxels are missing from the mapping.
Impact
In simulation workflows, this causes some concentration fields not to be remapped to the next frame, leading to signal loss over time.
Additional notes
• No obvious issues were noticed in the reassigned TIFF that tracks from the beginning to the end of the movie.
• This suggests the problem may be specific to the voxel_matches.npy output rather than the broader tracking result.