Unselfie: Translating Selfies to Neutral-pose Portraits in the Wild

  • 2020-07-29 19:21:02
  • Liqian Ma, Zhe Lin, Connelly Barnes, Alexei A. Efros, Jingwan Lu
  • 225

Abstract

Due to the ubiquity of smartphones, it is popular to take photos of one'sself, or "selfies." Such photos are convenient to take, because they do notrequire specialized equipment or a third-party photographer. However, inselfies, constraints such as human arm length often make the body pose lookunnatural. To address this issue, we introduce $\textit{unselfie}$, a novelphotographic transformation that automatically translates a selfie into aneutral-pose portrait. To achieve this, we first collect an unpaired dataset,and introduce a way to synthesize paired training data for self-supervisedlearning. Then, to $\textit{unselfie}$ a photo, we propose a new three-stagepipeline, where we first find a target neutral pose, inpaint the body texture,and finally refine and composite the person on the background. To obtain asuitable target neutral pose, we propose a novel nearest pose search modulethat makes the reposing task easier and enables the generation of multipleneutral-pose results among which users can choose the best one they like.Qualitative and quantitative evaluations show the superiority of our pipelineover alternatives.

 

Quick Read (beta)

loading the full paper ...