Monocular 3D Face Reconstruction with Joint 2D and 3D Constraints

3D face reconstruction from a single image is challenging problem, especially under partial occlusions and extreme poses. This because the uncertainty of estimated 2D landmarks will affect quality reconstruction. In this paper, we propose novel joint optimization method to adaptively reconstruct shapes image, which combines depths solve uncertain detections invisible landmarks. The strategy our involves two aspects: coarse-to-fine pose estimation using both landmarks, an adaptive re-weighting based on refined parameters recover accurate faces. Experimental results multiple datasets demonstrate that can generate high-quality color robust for self-occlusions large