Makeup studies have recently caught much attention in computer version. Two of the typical tasks are makeup-invariant face verification and makeup transfer. Although having experienced remarkable progress, both tasks remain challenging, especially encountering data in the wild. In this paper, we propose a disentangled feature learning strategy to fulfil both tasks in a single generative network. Overall, a makeup portrait can be decomposed into three components: makeup, identity and geometry (including expression, pose etc.). We assume that the extracted image representation can be decomposed into a makeup code that captures the makeup style and an identity code to preserve the source identity. As for other variation factors, we consider them as native structures from the source image that should be reserved. Thus a dense correspondence field is integrated in the network to preserve the geometry on a face. To encourage delightful visual results after makeup transfer, we propose a cosmetic loss to learn makeup styles in a delicate way. Finally, a new CrossMakeup Face (CMF) benchmark dataset (https://github.com/ly-joy/Cross-Makeup-Face) with in-the-wild makeup portraits is built up to push the frontiers of related research. Both visual and quantitative experimental results on four makeup datasets demonstrate the superiority of the proposed method.