Why not choose a uniform offset

[kai-lan]

If I understand everything correctly about AMASS dataset, offsets are the same for all subjects with the same body type ($\beta$ parameters).

As shown in https://github.com/KosukeFukazawa/CharacterAnimationTools/blob/main/anim/amass.py, offsets can be easily computed from the first frame of an animation.

If that is the case, can we just use the offsets derived from zero $\beta$ parameters, and keep the joint rotations as they are for each animation? That way, we can just do a fk to compute global positions, etc, and no need for ik.