MonoNPHM: Dynamic Head Reconstruction from Monocular Videos

Figure 1 . We present MonoNPHM, a neural-field-based parametric head model (left), for dynamic 3D head reconstruction in monocular videos (right). On the left we demonstrate the effect on a reconstructed human head, by individually varying shape (top box), expression (middle box) and appearance (bottom box). The right-hand side illustrates three input RGB frames (top row), and our reconstructed geometry (bottom row). We also show the reconstructed appearance under estimated lighting conditions, which is the basis of our reconstruction.