Joint Generative Image Deblurring Aided by Edge Attention Prior and Dynamic Kernel Selection

Figure 4

Visual effects of different methods on GOPRO: (a) blurred image and results of (b) DeblurGAN, (c) DMPHN, (d) SIUN, and (e) ours. The left images are global deblur results, while local restoration details are shown on the right. Our results show clear object boundaries without artifacts and produce various generative edge maps for the discriminator to judge the realness of the generation. The small zoom-in pictures of (e) show the good visual effect of edge attention prior and dynamic kernel selection.

(a)

(b)

(c)

(d)

(e)