Examples of reconstruction: Comparison of ground truth (GT) images with reconstructed (rec) results of the autoencoder (AE) and single-pixel-decoder (SPD) networks for two different compression ratios cr = 6.25% and cr = 1.56%. The first line shows the reconstructed images, whereas the second line shows the difference image Image_rec − Image_GT. Generally, fine details get neglected for lower compression ratios. Therefore, only the shapes of objects with fine details like e.g. greenery (see first line) get reconstructed. Objects with less high frequency components like traffic signs (last two examples) get reconstructed near perfectly even for the lower compression ratio.