For the map comparison, 50 possible maps have been collected using the search term ‘wallpaper’ in Google’s image search. This search term was used, since it is (i) a general term, without any specific image categories, (ii) wallpapers are likely to have a high resolution, and (iii) wallpapers are likely to be visually pleasant since they are often used as desktop backgrounds. The criteria for the images were that their minimum resolution was \(1920\times 1080\). Images with a higher resolution were converted to \(1920\times 1080\). For each image, we generated 1000 random image patches using the simulation method described in Section \ref{sec:draug}, followed by the histogram extraction method. This yielded a labeled dataset of histograms and corresponding positions. For each map, we determined the expected overall loss based on the method described in Section \ref{sec:syntheticdatageneration}. The compared maps were then sorted according to their estimated loss.