A problem about strand curvatures.

[XiaokunSun]

Hi! I noticed that you mentioned in neuralhaircut that the incorporation of curvature loss can better model curly strands. How do you define curvature? My understanding is that curvature represents the rate of change of orientation, and the implementation code is similar to:

# strand.shape = [N, 100, 3]
strand_ori = strand[:, 1:, :] - strand[:, :-1, :]
strand_ori= strand_ori/ (torch.norm(strand_ori, dim=-1)[:, :, None] + 1e-20)
strand_cur= strand_ori[:, 1:, :] - strand_ori[:, :-1, :]

Do I understand correctly? Looking forward to your reply