Supplementary Material for Uncertainty-Driven Loss for Single Image Super-Resolution

Neural Information Processing Systems 

The training cost mainly depends on the original networks and iterative times. The visual image comparison results on BI degradation are reported in Figure 1, Figure 1 and Figure 1. From Figure 1, we can see that our SR-resolved result of'Img_046' from Urban100 is recovered with From Figure 1, we can see that our SR result (Figure 1 (d)) of'Img_013' achieves the Single image super-resolution from transformed self-exemplars.