how can i use smpl rot6d motion_representation for motion reconstruct

I tried using humanact12 pretrain_model with the following command
"--cfg" ,"./configs/config_mld_humanact12.yaml", "--cfg_assets", "./configs/assets.yaml", "--task", "reconstrucion"
But an error is reported when calling this function "self.feats2joints" which is defined in mld.py line 132,because a parameter ”mask“ is missing in mld.py line 285.

Hi, thanks for your attention. The line 285 of mld.py does not include "mask" parameters.

motion-latent-diffusion/mld/models/modeltype/mld.py

Line 285 in af507c4

joints = self.feats2joints(feats_rst.detach().cpu())

If you want to use smpl rot6d, you need to remove this self.feats2joints. It is used to convert features into xyz joints.
If you are familiar with SMPL, you can use its joint regressor to replace self.feats2joints directly.

Thank you for your reply！I see, if I just want the result in SMPL rot_6d form, this "feats_rst = self.vae.decode(z, length)" in line 282 is all i need.