Unsupervised Object Representation Learning using Translation and Rotation Group Equivariant VAE