search

masabdi / LSPS

Source code for "3D Hand Pose Estimation using Simulation and Partial-Supervision with a Shared Latent Space"
https://arxiv.org/abs/1807.05380
GNU General Public License v3.0
61 stars 17 forks source link
hand-pose-estimation semi-supervised-learning synthetic-data unsupervised-learning

readme

Code for our BMVC oral paper (4.3% acceptance rate): "3D Hand Pose Estimation using Simulation and Partial-Supervision with a Shared Latent Space" see the paper at https://arxiv.org/abs/1807.05380

Citation

If you found this research useful, please cite:

@article{abdi20183d,
title={3D Hand Pose Estimation using Simulation and Partial-Supervision with a Shared Latent Space},
      author={Abdi, Masoud and Abbasnejad, Ehsan and Lim, Chee Peng and Nahavandi, Saeid},
      journal={arXiv preprint arXiv:1807.05380},
      year={2018}
}

Supplementary Video:

Real-time 3d hand pose estimation on CPU

Discriminative Results:

Alt text

Generative Results:

Alt text

Usage

  1. Use pose_train to train the vae: 

    python depth_train.py --config ../exps/nnyu.yaml

  2. Pretrain the depth model using: 

    python depth_train.py --config ../exps/nnyu.yaml --mode pretrain

  3. Finally run this command for the unsupervised setting: 

    python depth_train.py --config ../exps/nnyu.yaml --mode estimate3