Through "learning" from a large number of data sets, a properly trained neural network can "de-multiplex" high-resolution, spectrally crosstalk-free phases from the composite fringe and directly ...