Intelligent layout generation based on deep generative models: A comprehensive survey

Layout generation, as a crucial stage in Artificial Intelligence-Generated Content (AIGC) task information presentation generation, is a fascinating and significant research topic. Automatically generating layouts intrigued the computer vision community dating back to the 1990s, and the advancement was sluggish for a time due to the inherent complexity of the issue. Thanks to the success of deep learning, layout generation is leading to a surge in both academic literature and business applications recently. Yet a systematic and comprehensive literature review of this domain is still missing. In this study, we provide a comprehensive overview of the state-of-the-art layout generation methods. Primarily, we discuss the problem of layout generation and associated background knowledge including layout representation, layout aesthetics, and generative models. Subsequently, recent advances in the field of intelligent layout generation are organized and summarized from three perspectives: innovations, models & algorithms, and applications. Furthermore, the paper outlines the datasets and evaluation methodology, both qualitative and quantitive, and briefly analyzes the selection of them for specific tasks. To sum up, we conclude the current challenges of intelligent layout generation and suggest possible directions for future endeavors.