We propose a two-stage model that separates the in- painting problem into structure prediction and image com- pletion. Similar to sketch art, our model first predicts the image structure of the missing region in the form of edge maps. Predicted edge maps are passed to the second stage to guide the inpainting process. We evaluate our model end- to-end over publicly available datasets CelebA, CelebHQ, Places2, and Paris StreetView on images up to a resolution of 512 x 512. We demonstrate that this approach outper- forms current state-of-the-art techniques quantitatively and qualitatively.
The proposed network is able to use color and edge information from different sources to create unique images.
The proposed method can also be used for photo editing as seen below.
More information is available here.
For technical details please look at the following publications