Reposing and retargeting unrigged characters with intrinsic-extrinsic transfer

In the 3D digital world, deformations and animations of shapes are fundamental topics for several applications. The entertainment industry, virtual and augmented reality, human-robot interactions are just some examples that pay attention to animation processes and related tools. In these contexts, researchers from several communities desire to govern deformations and animations of 3D geometries. This task is generally very complicated because it requires several skills covering different kinds of knowledge. For this reason, we propose a ready-to-use procedure to transfer a given animation from a source shape to a target shape that shares the same global structure. Our method proposes highly geometrical transferring, reposing, and retargeting, providing high-quality and efficient transfer, as shown in the qualitative evaluation that we report in the experimental section. The animation transfer we provide will potentially impact different scenarios, such as data augmentation for learning-based procedures or virtual avatar generation for orthopedic rehabilitation and social applications.