/Capsule-HandNet

Primary LanguagePython

Capsule-HandNet

Installation

Install h5py for Python:

  sudo apt-get install libhdf5-dev
  sudo pip install h5py

-- Install Chamfer distance package:

  cd models/nndistance
  python build.py install

Usage

Data download

Download the MSRA Hand Gesture database at: https://www.dropbox.com/s/bmx2w0zbnyghtp7/cvpr15_MSRAHandGestureDB.zip?dl=0

Unzip it and put folders "P0"~"P9" in the "data/cvpr15_MSRAHandGestureDB" directory.

Please cite "Xiao Sun, Yichen Wei, Shuang Liang, Xiaoou Tang and Jian Sun. Cascaded Hand Pose Regression. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015", if you use this database.

Data preprocess

cd preprocess
matlab preprocess.m

Please cite this paper if you use this code. "Liuhao Ge, Yujun Cai, Junwu Weng and Junsong Yuan. Hand PointNet: 3D Hand Pose Estimation using Point Sets. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2018."

Train and evaluation

Pretrained Model trained on "P1"~"P9": BaiduNetDesk Link code：z8z8

put encoder model in 'AutoEncoder/results/P0' and regression model in 'results/P0'.

AutoEncoder Training and evaluating

cd AutoEncoder
python train.py
python eval.py

set "capsule model" as the saved model.

Hand pose regression

python train.py
python eval.py

References

Our capsule model is based on 3D-point-capsule-networks..