1aa3d9c6ce672447e1e5d0f1b5207e85-AuthorFeedback.pdf

We thank the reviewers for raising this question. SOTA methods are actively using such a parametrization (e.g., Gemici et al. (2016)) and we believe it is important12 toshowand evaluate this aspect. Moreover, please note that our model can straightforwardly be applied on higher dimensional manifolds.18 We thank (R1) for suggesting to expand on the limitations of our method. As28 reminded by (R2), projections required by our approach can increase the computational cost. This cost can be further reduced by only projecting the output of the ODE solver steps.