This work proposes a deep neural network approach for the phase retrieval problem, which learns a filtered spectral initialization. Simulation results show that the proposed approach requires fewer snapshots and power iterations than state-of-the-art.