A question for train_mesh.py

Walter0807 / MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"

Apache License 2.0

1k stars 120 forks source link

A question for train_mesh.py #105

Closed lllluy closed 9 months ago

lllluy commented 9 months ago

Hi I'm trying to retrain a human mesh model using my own dataset, and I've encountered some difficulties.

While examining the "mesh_det_h36m.pkl" data you provided, I noticed there are eight fields, and I'm particularly interested in understanding how "smpl_pose" and "smpl_shape" are computed. Could you please explain how these values are calculated?

Alternatively, if I already have "joint_2d" and "joint_cam" data, how can I compute the corresponding "smpl_pose" and "smpl_shape"?

I appreciate your assistance. Looking forward to your reply.

ShirleyMaxx commented 9 months ago

Hi, The smpl_pose and smpl_shape are SMPL parameters, which are provided officially by the Human3.6M dataset.

If you already have joint_2d and joint_cam data, you may consider using SMPLify or other recent works to optimize and get pseudo-GT SMPL parameters.