tf_smpl

TensorFlow interface for SMPL model. Code by https://github.com/akanazawa/hmr extracted as a separate repo.

Requires the SMPL model (.pkl file format):

wget https://people.eecs.berkeley.edu/~kanazawa/cachedir/hmr/models.tar.gz && tar -xf models.tar.gz

The downloaded SMPL model contains chumpy arrays. preprocess.py (adapted from https://github.com/CalciferZh/SMPL) converts these to numpy arrays. Running it requires the chumpy library which is provided with the SMPL model (http://smpl.is.tue.mpg.de/).

Citation

@inProceedings{kanazawaHMR18,
  title={End-to-end Recovery of Human Shape and Pose},
  author = {Angjoo Kanazawa
  and Michael J. Black
  and David W. Jacobs
  and Jitendra Malik},
  booktitle={Computer Vision and Pattern Regognition (CVPR)},
  year={2018}
}