GAN-Based Image Super-Resolution with a Novel Quality Loss

Figure 5

Visual comparisons for 4x SR on samples from BSD100 (PI/SSIM/PSNR(dB)). In general, GMGAN performs well not only in PI but also in visual effects.