Abstract
Arbitrary style transfer is the problem of synthesizing content image withstyle of the image that have never seen before. Recent arbitrary style transferalgorithms have trade-off between the content structure and the style patterns,or maintaining the global and local style patterns at the same time isdifficult due to the patch-based mechanism. In this paper, we introduce a novelstyle-attentional network (SANet), which efficiently and flexibly decorates thelocal style patterns according to the semantic spatial distribution of thecontent image. A new identity loss function and a multi-level featuresembedding also make our SANet and decoder preserve the content structure asmuch as possible while enriching the style patterns. Experimental resultsdemonstrate that our algorithm synthesizes higher-quality stylized images inreal-time than the state-of-the-art-algorithms.