Cascading and Residual Connected Network for Single Image Superresolution

Figure 11

Comparable training results (PSNR) with scale factor ×2 under variable dilations; blue result illustrates the training process without dilation while red one shows deformable convolution with dilation 2.