(d) Image inpainting results of the proposed approach. Edges drawn in black are computed (for the available regions) using Canny edge detector whereas edges shown in blue are predicted by the Edge Generator Network. Region shown in white are generated (for the masked regions) using a Mask Predictor Network. The missing regions are depicted in Black.
As a result, semantically realistic and visually appealing image is generated. This predicted edge map as well as an incomplete color image are sent to the refinement network for Image Inpainting. The resulting mask will then be used to predict the complete edge map using "Gated Convolutions". Initially, an "Autoencoder" Mask Prediction network gets an incomplete color image as input and generates masks. As our network architecture, we proposed a robust pipeline. This work proposes a novel method to solve image inpainting task without the need for any explicit mask. Passing an explicit mask to solve the image inpainting problem is therefore tricky. For example, naturally occurring Image deformations are random and do not have a defined mask. However, having accurate masks is difficult in practice in a Generative Image Inpainting using Edge Learning wihout a Maskĭeep learning approaches have resulted in considerable improvements in image inpainting during the last few years.
0 kommentar(er)
