Closed zhixiongzuo closed 2 years ago
Hello, I think directly choosing one of the 20 shape parameters is OK because the body shape of the same person does not change much under different poses. Suppose you expect to minimize the SMPL fitting errors brought by choosing one shape parameter. In that case, you can minimize vertex errors between two SMPL models generated from the original and chosen shape parameters by optimizing the pose parameters.
Thanks for your timely reply, I got it. By the Way, do you suggest me to use EasyMoCap multi view 1 prson to do smpl fitting, because the output smpl params like Rh and Th are not standard smpl parms, have you occur this problem ?
Please refer to their instructions on the output, then you can convert the output to pose vectors.
fine, thanks again !
Thanks for your working .In your paper, you said fisrst you fit smpl to each scan using EasyMoCap, and then defom the scans to canonical pose . I notice that there are 20 scans in example training data, as well as 20 pose txts, but there are only one shape txt. When I process my own data, after smpl fitting to each scan , there are 20 pairs of pose and shape params in which shape params are not same, how can I compute 1 shape params from these 20 shape params like your training data ?Or how did you fit multi scans to get animatable avatar?