Learning Complete 3D Morphable Face Models From Images and Videos

Figure 1. We present a method for learning complete 3D morphable models of faces from videos and images. We show visualizations of the learned models on the right. Faces in each direction of indicated arrows is obtained by linearly scaling individual component of respective models. Identity geometry captures variations in the face shape (second column), lips (top left to bottom right) and jaw (top right to bottom left), while expressions capture variations due to mouth opening (second row), smile (second column) and eye movement (top right to bottom left). Albedo/Reflectance spans a variety of skin color (second column), eye color (top right to bottom left) and gender specific features such as facial hair and make-up (second row).