6d rotation representation

pytorch batch implementation with batch geodesic loss which can be integrated inside a network.

As described in https://arxiv.org/abs/1812.07035.

On the Continuity of Rotation Representations in Neural Networks. Yi Zhou et. al