We share the project page of “Learning Character-Agnostic Motion for Motion Retargeting in 2D”, the paper published at SIGGRAPH 2019.
In this paper, we created a deep learning model that extracted motion, skeleton, and camera angle from three images (which may differ from each other) and then combined them into one. Individual or combination retargeting for Motion, Skeleton, and Camera Angle is possible. Mixamo was used as training data.
The author says that the result can be used in various fields such as video performance cloning as well as motion retrieval. I plan to use it for 3D or 2D character animation in the future.