smplify-x
smplify-x copied to clipboard
Question about making SMPL ground truth
Thank you for sharing the great work. I have met some questions when using your method.
Here is a picture (s_01_act_02_subact_01_ca_01_000001) of the visualization of the produced human 3d joint coming from smpl model and human3.6M ground truth.
Though impressive result has been achieved, I still have some questions.
If I want to make the dataset of my own by using your method. What is the camera paras and trans paras like? For example, your method will produce camera_rotation
(identity matrix), camera_translation
, betas
, global_orient
, body_pose
(32-dim), pose
(69-dim). But the camera_rotation
and transl
for other datasets with smpl values (like 3DPW, Human3.6M) are not identity matrix or none. I am confused with the Figure_2 because I think there may be a rotation transformation missed for the smpl model (otherwise the two skeletion will match perfectly). The paras produced by your method and other method for s_01_act_02_subact_01_ca_01_000001 are list as below,
# your method
camera_rotation = [[[1. 0. 0.], [0. 1. 0.], [0. 0. 1.]]]
camera_translation = [[-0.12283573 -0.12966573 26.40824 ]]
betas = [[ 9.0937483e-01 1.4290957e+00 -4.9474990e-01 -1.7118127e+00, -1.8398558e-01 -1.6947743e-01 4.5224284e-03 3.0939683e-01, -8.9674111e-05 1.4452808e-01]]
global_orient = [[ 0.54963225 0.23515686 -3.0548875 ]]
body_pose = ...
pose = ...
Thank you.