Surprising Image Compositions

Othman Sbai, Camille Couprie, Mathieu Aubry; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 3926-3930

Visual metaphors are a powerful and effective way of communication in advertising, news and art. By taking objects out of their natural context, artists create new and surprising composite images by leveraging visual, linguistic or phonetic analogies. We build on recent image retrieval, completion and composition methods to design a new collage generation tool with the aim of assisting artists in creating interesting composite images. Given a selected object in an image, our model searches for visually similar but semantically different objects and performs the image blending automatically, leading to surprising image combinations. Using automatic metrics and a human study, we test our approach against improved baselines and show the potential of this novel artistic application.