Jia-Wang Bian, Huangying Zhan and Ian Reid

We present a novel method to synthesize photo-realistic novel view selfies from a mobile phone captured video. This is challenging due to the inconsistent geometry, which is caused by the dynamic person. Recent methods reconstruct the whole deformable scene implicitly with a deformation field. We argue that they are inefficient and hard to fit diverse real-world videos. Instead, we use an explicit reconstruction for generalization and efficiency, and we separately track, reconstruct, and synthesize the foreground and background to overcome the geometry inconsistency. Several novel and effective modules are proposed for better performance and visual quality. We demonstrate the advantage of the proposed method against the existing alternatives in a collection of our captured selfie videos with the support of quantitative and qualitative results.