SICON (Stokes Inversion based on COnvolutional Neural networks) provides a three-dimensional cube of thermodynamical and magnetic properties from the interpretation of two-dimensional maps of Stokes profiles by use of a convolutional neural network. In addition to being much faster than parallelized inversion codes, SICON, when trained on synthetic Stokes profiles from two numerical simulations of different structures of the solar atmosphere, also provided a three-dimensional view of the physical properties of the region of interest in geometrical height, and pressure and Wilson depression properties that are decontaminated from the blurring effect of instrumental point spread functions.

torchmfbd carries out multi-object multi-frame blind deconvolution (MOMFBD) of point-like or extended objects, and is especially tailored to solar images. The code is built on PyTorch and provides a high-level interface for adding observations, defining phase diversity channels, and adding regularization. It can deal with spatially variant PSFs either by mosaicking the images or by defining a spatially variant PSF. torchmfbd supports smooth solutions and solutions based on the ℓ 1 penalization of the isotropic undecimated wavelet transform of the object, and regularizations are easily extendable. The code also includes an API and a configuration file.