Mapping galaxy images across ultraviolet, visible and infrared bands using generative deep learning

We demonstrate that generative deep learning can be used to map astronomical observations of galaxies across various photometric bands including ultraviolet visible and infrared bands. We also demonstrate that this mapping can be learned without the need to use existing multi wavelength observations which are sparse and nontrivial to obtain and preprocess. We do so by using mock observations from the Illustris cosmological simulations to train our models. To date, this is a novel use of Illustris’ mock observations. We demonstrate that models trained on mock observations can generalize and extend the learned mapping to real observations using general image comparison metrics such as the structural similarity index (SSIM) and more specific astronomical metrics such as the GINI coefficient and the M20 measurement. This could allow astronomers to significantly augment existing observations, specifically in areas where multiwavelength observations do not exist, with minimal inference cost.