Image completion with inpainting using deep learning

In this recipe, you will learn how a Fully-Convolutional deep learning Network (FCN, called a Completion Network model, from a recent paper, Globally and Locally Consistent Image Completion from SIGGRAPH 2017) can be used to complete the missing parts of a (previously unseen) image. We shall specifically use a pre-trained version of the neural network model to predict the missing parts in an image. The model will accept an input image and a mask (corresponding to the missing parts in the image) and try to predict the missing parts from the information provided by the remaining (incomplete) image.

Getting ready

Download the pre-trained torch model for Globally and Locally...