To run the registration scripts, different files are required depending on whether you use SMPL or SMPL-H model:
|model root
|--SMPL_male.pkl # SMPL body male model, required if you use SMPL
|--SMPL_female.pkl
|--SMPL_neutral.pkl
|--SMPLH_female.pkl # SMPLH body models, required if you use SMPL-H
|--SMPLH_male.pkl
|--priors # folder containing body and hand pose priors
|----body_prior.pkl # body prior, required both in SMPL and SMPL-H
|----lh_prior.pkl # hand priors, required if you use SMPL-H model
|----rh_prior.pkl
|--regressors # folder for body, face, and hand regressors, required both for SMPL and SMPL-H
|----body_25_openpose_joints.pkl
|----face_70_openpose_joints.pkl
|----hands_42_openpose_joints.pkl
For SMPL body model files SMPL_*.pkl
, you can download from the SMPL website, once download and unzip finished, you can rename the model files in the following way to follow our convention:
basicmodel_f_lbs_10_207_0_v1.1.0.pkl -> SMPL_female.pkl
basicmodel_m_lbs_10_207_0_v1.1.0.pkl -> SMPL_male.pkl
basicmodel_neutral_lbs_10_207_0_v1.1.0.pkl -> SMPL_neutral.pkl
For SMPL-H body model files SMPLH_*.pkl
, you can download from the SMPL-H website and place them to your model root accordingly.
You can download our prebuilt priors in assets/priors
. The body prior body_prior.pkl
is required whether you use SMPL or SMPL-H. The hand priors lh_prior.pkl
and rh_prior.pkl
are required if you use SMPL-H model.
The body prior was built from a subset of AMASS dataset and hand priors were built from GRAB dataset.
Alternatively you can build your own priors from another dataset using the script utils/build_prior.py
:
python utils/build_prior.py data_path out_path
You can download these files from assets/regressors
.
Once these files are ready, you can change SMPL_MODELS_PATH
in config.yml
accordingly.