Image extrapolation based on multi-column convolutional attention network

Image inpainting is based on the generation of branches against the network. Many recent methods of deep learning have shown great progress in challenging tasks that repair large numbers of missing areas in the map. These methods can be visually restored to a reasonable image and structure, but still produce a distorted structure or a fuzzy structure that is different in the surrounding area. In the past year, image expansion has begun to slowly enter our field of vision. Previous research based on mathematical methods started to get noticed, but the challenge of extrapolating this work is very huge, so we propose multi-column convolution. The attention network is trained to produce a blank gap in the image, and the depth learning method of image expansion is very promising and proved to be feasible.