论文: Deep Feature Interpolation for Image Content Changes
can perform high-level semantic transformations like “make older/younger”, “make bespectacled”, “add smile”, among others, surprisingly well—sometimes even matching or out-performing the state-of-the-art.
attribute bias across the two data sets. If for example, all target images in St were images of more senior people, and source images in Ss of younger individuals the vector wwould also capture the change involved in aging. Also, if the two sets are too different from the test image (e.g., a different race) the transformation would not look believable. To ensure sufficient similarity we restrict St and Ss to the Knearest neighbors. Let NKt denote the K nearest neighbors of St to φ(x) and we define
论文核心内容
本文由zdx3578推荐。