Abstract: Computer vision advancements allow motion transfer for animating static objects in images. However, current methods rely on manually collected motion labels and struggle with accurate shape ...