Data consistency class can be used for: -performing E^h*E operation in the paper -transforming final network output to kspace

Parameters ---------- input_x: batch_size x nrow x ncol x 2 sens_maps: batch_size x ncoil x nrow x ncol trn_mask: batch_size x

Parameters ---------- rho: split ratio for training and loss mask. \ rho = |\Lambda|/|\Omega| small_acs_block: keeps a small acs reg