Learning Viewpoint-Agnostic Visual Representations by Recovering Tokens in 3D Space - Citation Graph | Papersgraph