Question about training shape and pose parameters
duyguceylan opened this issue · comments
Very nice work and thanks for sharing your code. After reading your paper, I have a question regarding the training of the STAR model. The original SMPL paper first trains the pose corrective deformation functions and utilizes these to "normalize" the pose of the 3D scans before learning the shape PCA space. However, since the pose corrective deformations for STAR also depend on the beta_2 shape parameter, I wonder what the order of training was. Did you estimate beta_2 from the BMI annotations (as you show that BMI and beta_2 are correlated) and first trained the pose correctives? If not, did you first train the shape space and how did you do the pose normalization? Or do you do joint/iterative training?