In millimeter wave (mmWave) systems adopting large number of antennas at both the transmitter and the receiver, channel estimation is challenging due to the large number of antennas and low signal-to-noise ratio (SNR) before beamforming. Due to the sparse nature of mmWave channels, the channel estimation can be solved by beam search. While conventional beam search schemes rely on exhaustive training, it requires excessive overhead scaling proportional to the number of antennas. Compressed sensing (CS) is a promising approach to reduce the training overhead. However, CS shows poor recovery performance with low SNR. In this paper, we propose an asymmetric channel estimation scheme that combines an exhaustive beam training at the receiver side and a CS approach for transmit beam training. The hybrid scheme shows robust performance in a low SNR regime and can be applied for multi-user systems.