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

Figure 2

MRF framework. The image is separated into different scales from top to bottom. (a) The extraction path of extracting features from scales. (b) Fusion of the recurrent last-round results and the upsampling feature maps as a single refinement process. All four refinement paths finally compute the loss in the scale refinement loss function, and then, the best deblur results are obtained.