Unsupervised Object-Centric Learning from Multiple Unspecified Viewpoints

Visual scenes are extremely diverse, not only because there are infinite possible combinations of objects and backgrounds but also because the observations of the same scene may vary greatly with the change of viewpoints.