SMPLX_Parser Issue with AMASS Data Conversion

[www-Ye]

I noticed that when processing AMASS data into SMPLX format using SMPLX_Parser, it seems necessary to comment out lines 421-422 in the smpl_parser.py code to run properly:

th_betas = torch.cat([th_betas, torch.zeros((B, 20 - beta_shape)).to(th_betas)], dim = -1)

Changing the dimension to 20 seems to prevent the SMPLX part from executing correctly, showing a mismatch in the corresponding dimensions of shapedirs and shape_components. What could be the reason for this?

I am using the models_smplx_v1_1.zip model file, and the SMPLX code is from the main branch of https://github.com/vchoutas/smplx.