Zero-Shot Image Harmonization with Generative Model Prior

arXiv:2307.08182v2 Announce Type: replace
Abstract: We propose a zero-shot approach to image harmonization, aiming to overcome the reliance on large amounts of synthetic composite images in existing methods. These methods, while showing promising results, involve significant training expenses and often struggle with generalization to unseen images. To this end, we introduce a fully modularized framework inspired by human behavior. Leveraging the reasoning capabilities of recent foundation models in language and vision, our approach comprises three main stages. Initially, we employ …

abstract arxiv cs.cv generative image images prior reliance results struggle synthetic training type zero-shot