A Shape-Aware Retargeting Approach to Transfer Human Motion and Appearance in Monocular Videos

Transferring human motion and appearance between videos of actors remains one the key challenges in Computer Vision. Despite advances from recent image-to-image translation approaches, there are several transferring contexts where most end-to-end learning-based retargeting methods still perform poorly. actor to another is only ensured when a strict setup has been complied, which generally built considering their training regime’s specificities. In this work, we propose shape-aware approach based on hybrid image-based rendering technique that exhibits competitive visual quality compared state-of-the-art neural approaches. The formulation leverages user body shape into while physical constraints 3D 2D image domain. We also present new video benchmark dataset composed different with annotated motions evaluate task synthesizing people’s videos, can be used as common base improve tracking progress field. its evaluation protocols designed more general challenging conditions. Our method validated experiments, comprising publicly available shapes, types, camera setups. code community at: https://www.verlab.dcc.ufmg.br/retargeting-motion.