Table 1 Scores computed for the 30 GAN-generated 2563 images using the full multi-scale pipeline. We refer to (2) and (1) for details. The reference is computed by comparing two sets of 15 real samples. It represent the best score that can be achieved