Conversion to MHR parameters

Hi, thanks a lot for the study and the code!

I was wondering if it was possible to convert resulting mesh representation to MHR parameters since its topology is MHR LOD6. I'm trying to do ICP with and without the assumption of point to point correspondence, by gradually unlocking the global orient + translation (with centroid alignment), poses, betas, and regularized betas + velocity + acceleration.

Didn't want to go down the route of converting to SMPL-X first (which the repo already does natively), and then to MHR, since it will compound the errors along the way, at least that was my chain of thought. And the resulting mesh's topology of the paper was identical to MHR LOD6, so it seemed like the best way to do this.

Is there an obvious way to do this that I'm missing? Or maybe what I'm doing is not necessary / recommended at all?

Thanks a lot again!

==========

After opening the issue I realized the topology differs from the MHR LOD6, not a big difference but different enough to degrade performance. So the point to point correspondences assumption does not hold. It was my misunderstanding during reading the paper.

The issue names the repository's native SMPL-X conversion path and MHR LOD6 topology; start by reviewing that conversion path and checking how its mesh topology is compared with MHR. Done would be a documented recommendation on whether direct MHR-parameter conversion is supported or appropriate, including the correspondence limitation described in the issue.

Written by the indexing model from the issue text.