Atlases are critical for mapping sample data into the correct anatomical space for quantification and comparison across multiple samples. Atlases across species and organs have been generated typically by serially labeling 2D planes, a time-consuming and inherently error-prone task that can lead to artifacts such as boundary misalignments between planes apparent in orthogonal dimensions. The growing body of whole organ volumetric microscopy has made the need for complete, accurate, and fully 3D atlases even more pressing, and variability across development mandates generation of such atlases specific to each age, a monumental task by manual methods. To generate such atlases while leveraging the large existing body of atlases, here we propose automated methods to extend atlases from serial 2D to fully 3D whole organ maps, even for only partially labeled organs. We employ 3D smoothing techniques to minimize region boundary artifacts in all dimensions while maintaining each region's overall shape and placement. To further refine labels, we automatically identify and incorporate gross anatomical boundaries into existing atlases to improve correspondence between labels and underlying anatomical markers, including those that may not have been as apparent along a single axis. Using the Allen Developing Mouse Brain Atlas series, we apply these methods to each atlas from prenatal to adult stages to demonstrate improved label coverage and anatomical correspondence across development as measured by distances between gross anatomical and label borders and within-region variability in 3D. Applying the atlas refinements to quantification of C57BL/6J mouse brains at age P0, we replicate these improvements on intact tissue cleared volumetric microscopy of mouse brains and whole brain nuclei detections including to show improved atlas alignment and regional quantification at both gross anatomical and cellular levels.