FEXNet: Foreground Extraction Network for Human Action Recognition

Overview

We release the PyTorch code of FEXNet.

The code is majorly based on TSM. The global structure of FEXNet is shown in the figure above. The proposed Foreground EXtraction blocks contains Scene Segregation and Foreground Enhancement modules to extract foreground features in different aspects as below.

Detailed structure of SS Module:

Detailed structure of FE Module:

The detailed data pre-processing and preperation strategies follow the settings of TSM.

Prerequisites

The code is built with following libraries:

• PyTorch 1.5.0 or higher
• TensorboardX
• tqdm
• scikit-learn

Training & Testing

The configuration of the super parameters also follows TSM.

Please refer to Training.sh and Testing.sh provided in the project for the detailed configuration.

Citation

@article{shen2021fexnet,
  title={FEXNet: Foreground Extraction Network for Human Action Recognition},
  author={Shen, Zhongwei and Wu, Xiao-Jun and Xu, Tianyang},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2021},
  publisher={IEEE}
}