Real-time 3D Face Super-resolution from Monocular In-the-wild Videos

A fully automatic approach to real-time 3D face reconstruction from monocular, 2D, in-the-wild videos is presented.

We use a low-resolution 3D Morphable Face Model to obtain a semi-dense shape and combine it with a fast median-based super-resolution technique to obtain a high-fidelity textured 3D face model. To select the most suitable frames to reconstruct the texture, this work employs a keyframe extraction that selects the best frames and vertices according to sharpness, pose angle and strength of expressions.

The system does not need prior training and is designed to work in uncontrolled scenarios.


This work will be presented at the SIGGRAPH 2016 poster session.

When: 24–28 July 2016

Where: Hall C

Anaheim, California


  • Abstract PDF (411 Kb) with an overview of the work.
  • Poster PDF (1.06 Mb) explains the details of the 3D shape-fitting, keyframe extraction and super-resolution.