Home -> Projects ->All Pose Face Alignment Robust to Occlusion Member Login  

Jump to: Demos | Publications | Releases


Previous methods are vulnerable to partial occlusion of the face, since they assumed, explicitly or implicitly, that no significant occlusion was occurred. Our approach relies on two schemes: one is discrete representation and utilization of multi-modal response from each of the face feature detectors, and the other is hypothesize-and-test search for the correct alignment over subset samplings of those in the feature response modes.
Based on the ideas above, we developed a robust all pose face alignment method regardless of partial occlusion.


Face alignment results on the CMU Multi-PIE. Each column represents camera angle.

Face alignment results on the LFPW.

Face alignment on various face images.
White curves and broken red curves represent regions which are determined as visible and occluded, respectively.

 Related Publications

 Releases [ Members Only ]