Radio Sky Models for Cosmodeep Training

A sample dataset of sky models and noise images for the training of the convolutional neuronal network "Cosmodeep" (Gheller et al. 2018), tailored to the analysis of ASKAP-like data. The dataset contains: a) 61 sky***.fits FITS files, 2000 x 2000 pixels each, with values in units of Jy/arcsec^2 and referred to the observing frequency of 1.4 GHz. b) 61 corresponding noise***.fits FITS files, containing the random noise level, tailored for the ASKAP-EMU survey. Each pixel is 10", and the image extend roughly corresponds to the ASKAP field of view. The sky models only contains diffuse synchrotron emission from electrons accelerated by cosmic shocks. Each sky***fits map is the result of the integrated emission of simulated boxes, from z=0.02 to z=0.5, simulated with the ENZO code (see Vazza et al. 2015 A&A for more details). The noise model only includes the estimated rms thermal noise for the EMU survey (~10 micro Jy/beam). See http://cosmosimfrazza.myfreesites.net/cosmodeep-training-datasets for more details.