Reconstructing Signing Avatars from Video Using Linguistic Priors

Figure 1 . Given a monocular, in-the-wild video of a sign-language sign, SGNify automatically reconstructs a 3D body with accurate hand pose, facial motion, and body pose. Note that motion blur obscures the finger articulations in several video frames; this is a common problem. Our novel linguistic priors enable accurate 3D reconstruction despite such image degradation.