Architecture Search for Image Inpainting

Abstract

Neural Architecture Search (NAS) shows the ability to automate the architecture engineering for specific tasks recently which is extremely promising. Many published works apply reinforcement learning or evolutionary algorithm to design the neural architecture for image classification and achieve state-of-the-art performance. However, using NAS to perform other challenging tasks, such as inpainting irregular regions in an image, has not been explored yet. The target of image inpainting is to generate plausible image regions to fill the missing regions in the original image. It has been widely used in many applications. In this paper, we are interested in applying neural architecture search methods to image inpainting tasks. We propose to use reinforcement learning to automatically design the network architecture. Our method can efficiently explore new network structure based on existing architecture. The experiment result demonstrates that the proposed method can design an efficient and high-quality architecture for image inpainting.