Self-Supervised 3D Face Reconstruction via Conditional Estimation

TL;DR

This paper introduces CEST, a self-supervised framework for 3D face reconstruction from 2D images using analysis-by-synthesis and statistical dependency modeling, trained on videos without explicit labels.

Contribution

The novel CEST framework estimates 3D facial parameters conditioned on previous estimates, leveraging symmetry and consistency for self-supervised learning from videos.

Findings

Effective 3D face reconstruction demonstrated on in-the-wild videos.

Improved disentanglement of facial parameters through symmetry and consistency constraints.

Qualitative and quantitative results validate the approach.

Abstract

We present a conditional estimation (CEST) framework to learn 3D facial parameters from 2D single-view images by self-supervised training from videos. CEST is based on the process of analysis by synthesis, where the 3D facial parameters (shape, reflectance, viewpoint, and illumination) are estimated from the face image, and then recombined to reconstruct the 2D face image. In order to learn semantically meaningful 3D facial parameters without explicit access to their labels, CEST couples the estimation of different 3D facial parameters by taking their statistical dependency into account. Specifically, the estimation of any 3D facial parameter is not only conditioned on the given image, but also on the facial parameters that have already been derived. Moreover, the reflectance symmetry and consistency among the video frames are adopted to improve the disentanglement of facial parameters. Together with a novel strategy for incorporating the reflectance symmetry and consistency, CEST can be efficiently trained with in-the-wild video clips. Both qualitative and quantitative experiments demonstrate the effectiveness of CEST.