This repository contains an example script to convert from a SMPL model to a bvh file.
The left side of the figure shows the SMPL grand truth and the right side shows the bvh data.
This code is MIT licensed, but SMPL requires a separate license.
Please see SMPL official website.
You need to download SMPL models in ./data/smpl/smpl/.
You need to download smplx too.
To install from PyPi simply run:
pip install smplx[all]
After downloads all requirements, you can use smpl2bvh like this:
python smpl2bvh.py --gender MALE --poses ${PATH_TO_Y0UR_INPUT} --fps 60 --output ${PATH_TO_SAVE} --mirrorposes is an .npz file or .pkl file.
.npz file must contain rotations and trans as keys.
rotations value is an np.array consisting of [fnum, 24, 3] and trans value is the root transition consisting of [fnum, 3]
(fnum means frame number).
.pkl file must contain smpl_poses and smpl_scaling and smpl_trans as keys.
smpl_poses value is an np.array consisting of [fnum, 72] and smpl_scaling value is the scaling parameter. smpl_trans value is the root transition consisting of [fnum, 3].
The format of pkl file is the same as AIST++ dataset.
If you check --mirror as an argument, the mirrored motion is also saved.
After processing, you can find bvh file as --output.
For more information, please refer to smpl2bvh.py.
bvh.py and quat.py are based on Motion Matching.