By not allocating the whole segmentation when saving or loading, but instead we stream the data, we can prevent memory overflow when allocating the huge 4D volume.
@Sunderlandkyl is working on more efficient handling of non-overlapping segment groups. If that works then we may not need streaming reading/writing after all.
By not allocating the whole segmentation when saving or loading, but instead we stream the data, we can prevent memory overflow when allocating the huge 4D volume.