2017/05/18 by Golnaz Ghiasi, Ghiasi, Golnaz, Honglak Lee +7 · 3 citations
Neuroscience · Computer Science · #Aesthetic Perception and Analysis #Generative Adversarial Networks and Image Synthesis #Computer Graphics and Visualization Techniques
paper · pdf · doi:10.48550/arxiv.1705.06830
In this paper, we present a method which combines the flexibility of the\nneural algorithm of artistic style with the speed of fast style transfer\nnetworks to allow real-time stylization using any content/style image pair. We\nbuild upon recent work leveraging conditional instance normalization for\nmulti-style transfer networks by learning to predict the conditional instance\nnormalization parameters directly from a style image. The model is successfully\ntrained on a corpus of roughly 80,000 paintings and is able to generalize to\npaintings previously unobserved. We demonstrate that the learned embedding\nspace is smooth and contains a rich structure and organizes semantic\ninformation associated with paintings in an entirely unsupervised manner.\n