Currently only tensor images in atlas space are computed for each stage, but not for the concatenated/final space (when computing a hierarchical atlas). For the QC of the generated atlas, we need to see the individual tensor images in the final atlas space, so these would need to be generated.