This dataset contains the input of the model for neural network with a UResNet architecture for application as surrogate model for the prediction of magnetic stray fields in two-dimensional random micro-heterogeneous materials. It consist of a colletion of 500 files of two-dimensional binary image data, in which a pixel with a value of 0 (1) corresponds to the magnetic solid (vacuum space), of the size of 128x128 pixel. As well included is short description of the applied configuration of the UResNet used in the related publication. The uploads are provided as exports from the shared electronic lab notebook (ELN) instance in .eln format (fully .zip archive compatible).