U-Net deep neural network used in the present work as an emulator for the spatially variant convolution. M′ = M + 1 is the number of channels of the input tensor, which is built by concatenating the image of interest and the spatially variant images of the M Karhunen-Loève. All green blocks of the encoder are conditioned to the specific instrumental configuration as described in Sect. 2.2.1 via the encoding fully connected network shown in the figure. The gray box refers to the one-hot encoding of the instrumental configuration.