Three-dimensional head pose estimation in-the-wild

Estimating 3-dimensional head pose from a single 2D image is a challenging task with extensive applications. Existing approaches lack the capability to deal with multiple pose-related and - unrelated factors in a uniform way. Most of them can provide only 1-dimensional yaw estimation and suffer from limited representation ability for out-of-sample testing inputs. These drawbacks limit their performance especially on faces in-the-wild. To address this problem, we propose a new head pose estimation approach, which models the pose variation as a 3-sphere manifold embedded in the high-dimensional feature space. It can uniformly factorize multiple factors in an instance parametric subspace, where novel inputs can be synthesized under a generative framework. Moreover, our approach can effectively avoid the manifold degradation issue by learning the embedding in a novel direction. The pose estimation results on multiple databases demonstrate the superior performance of our approach compared with the state-of-the-arts.