The Importance of Growing Up: Progressive Growing GANs for Image Inpainting

In recent years, Generative Adversarial Networks (GANs) have proven to be a sophisticated approach for generative tasks in image processing, especially inpainting and image synthesis While most GAN approaches feature comparatively large networks, we introduce an approach to image inpainting using progressive growing GANs, which enables significantly reduced model sizes, faster convergence, and thus an overall more efficient training that is inspired by insights on the development of visual abilities in biological systems. We demonstrate the effectiveness and efficiency of our approach on Places, a comprehensive dataset encompassing a wide variety of images from diverse locations in the wild.