Evolution of Modern Deep Image Inpainting Methods

Deep Image Inpainting is a field of research where damaged or deteriorated images are restored by adding the missing information or eliminating undesirable details using the assistance of deep learning techniques to make the final image relevant and visually appealing. Deep learning methods have evolved immensely over the past few years and now can work in extreme cases to produce semantically coherent images without any visible artifacts. This paper is an attempt to summarize some of the best algorithms and techniques with necessary comparisons by analyzing the pros and cons. In this paper, the previous works have been classified and arranged based on their network architecture thus presenting a comprehensive explanation for their approach and also highlights the issues that have a lot of room for improvement, such as training time, mask size and location, dealing with high resolution and diverse images, and so on.