A quick-and-dirty approach could be using MODELLER (the homology modeling software). MODELLER can work with just the C-alpha positions in the PDB and the sequence, and from there it can reconstruct an all-atom PDB. Just use your coarse grained PDB as a template.