Conv3D_CLSTM

The code of the new architecture "Res3D+ConvLSTM+MobileNet" implemented by TF-1.2 has been released in https://github.com/GuangmingZhu/AttentionConvLSTM, this repository is not in support now!

Prerequisites

1. Tensorflow-0.11
   
2. Tensorlayer (commit ba30379f1b86f930d6e86e1c8db49cbd2d9aa314)
   git clone https://github.com/zsdonghao/tensorlayer.git
   git checkout ba30379f1b86f930d6e86e1c8db49cbd2d9aa314
   

   The original Tensorlayer does not support the convolutional LSTM, so the class RNNLayer in the tensorlayer/layers.py needs to be modified according to the tensorlayer-rnnlayer.py.
   

Get the trained models

The trained models used in the paper can be obtained on the link: https://pan.baidu.com/s/1o8kDT9K Password: efm2.

How to use the code

Use training_.py to train the networks. Please replace the paths in the codes with your paths first.
Use testing_.py to validate the networks. Please replace the paths in the codes with your paths first.

Citation

Please cite the following paper if you feel this repository useful.

@article{Zhu2017MultimodalGR,
  title={Multimodal Gesture Recognition Using 3-D Convolution and Convolutional LSTM},
  author={Guangming Zhu and Liang Zhang and Peiyi Shen and Juan Song},
  journal={IEEE Access},
  year={2017},
  volume={5},
  pages={4517-4524}
}

Contact

For any question, please contact

  gmzhu@xidian.edu.cn