Deep generation of 3D articulated models and animations from 2D stick figures

Generating 3D models from 2D images or sketches is a widely studied important problem in computer graphics. We describe the first method to generate human model single sketched stick figure. In contrast existing modeling techniques, our does not require statistical body shape model. exploit Variational Autoencoders develop novel framework capable of transitioning simple figure sketch, corresponding Our network learns mapping between input sketch and output Furthermore, embedding space around these models. demonstrate that can only models, but also animations through interpolation extrapolation learned space. addition we produce horse order show generalization ability framework. Extensive experiments compatible with sketches. • generates given for sketch-3D mapping. any other such as rigged template