SCANimate: Weakly Supervised Learning of Skinned Clothed Avatar Networks

Figure 1 : SCANimate. Given a set of raw scans with multiple poses containing self-intersections, holes, and noise (left), SCANimate automatically aligns all scans to a canonical pose (middle) and learns a Scanimat, a fully animatable avatar that produces pose-dependent deformations and texture without garment-specific templates or mesh registration (right).