Confusion with the joints

[andypinxinliu]

I am confused with the smpl joints you have got and its difference with the smplh used in humanml3d. It seems like if you need to call TMR for evaluation, you did

x, y, z = T(motions)
motions = T(np.stack((x, z, -y), axis=0))

to convert the raw smpl joints in your smplh format to guo's format. May I know the reason why you need to have y to be negative? In the original guo's raw pose preprocessing, after getting the joints, his transform is

trans_matrix = np.array([[1.0, 0.0, 0.0],
                            [0.0, 0.0, 1.0],
                            [0.0, 1.0, 0.0]])

to let the joints multiply with this matrix, to make y and z axies to be switched. However, when you get the joints, not only did you switch two axies, but also make y to be negative, why is that?