Contextual Residual Aggregation

What is Contextual Residual Aggregation? Contextual Residual Aggregation, or CRA, is a state-of-the-art module used for image inpainting. The main function of the module is to fill in missing or damaged parts of an image with realistic and believable content. CRA produces high-frequency residuals for missing contents by weighted aggregating residuals from contextual patches, thus only requiring a low-resolution prediction from the network. Specifically, it involves a neural network to predict a