We have applied a combination of blind deconvolution and deep learning to the processing of Shack--Hartmann images. By using the intensity information contained in spot positions, and the fine structure of the separate images created by the lenslets, …