Electron microscopy data quite often have a problem with the estimated standard deviations of the amplitudes. Let me illustrate this with an example. The figure below compares the conventional and GraphEnt maps for a 8Å potential density projection of a large complex.
It looks as if all low resolution information disappeared from the GraphEnt map, and this is more-or-less what has indeed have happened. The reason is shown in the next figure. The two graphs show on the same scale the distribution of log10(F/(F)) versus resolution for the EM data (left graph) and of a typical X-ray crystallographic data set (right graph).
Whereas the X-ray data have a dynamic range extending approximately over two orders of magnitude, the EM data show a flat distribution with the (strong) low resolution terms having a value of F/(F) not much different from the data in the highest resolution shell. Because I have seen this behaviour with almost all EM data sets that I have come across, I suspect that the problem is with the data processing programs used by the EM community.