Lightweight Binocular Facial Performance Capture under Uncontrolled Lighting1.5ex]Supplementary Material1ex]

Fig. 1 shows the comparison of our structure-aware smoothness term with the TV regularizer used in [Valgaerts et al. 2010] for the target frame of Fig.6 in the paper. The top results are obtained using TV regularized scene flow and the bottom results using scene flow with our proposed anisotropic regularizer. The two zoom-ins clearly show that anisotropic regularized scene flow produces more realistic tracked features, while TV regularized scene flow induces drift artifacts, such as double folding and degenerate triangles, in particular in the mouth and eye brow region.