GTSNet: Flexible Architecture under Budget Constraint for Real-Time Human Activity Recognition from Wearable Sensor

Abstract

This program (GTSNet) is designed to perform the real-time sensor-based activity recognition. The GTSNet includes a Grouped Temporal Shift (GTS) module that allows the network architecture to be flexibly modified by predefining the theoretical computational cost.

This software is a PyTorch implementation of the proposed method. The original version of this program was written by Jaegyun Park.

[Paper]

@article{park2023gtsnet,
  title={{GTSNet}: Flexible architecture under budget constraint for real-time human activity recognition from wearable sensor},
  author={Park, Jaegyun and Lim, Won-Seon and Kim, Dae-Won and Lee, Jaesung},
  journal={Engineering Applications of Artificial Intelligence},
  volume={124},
  number={106543},
  year={2023},
  publisher={Elsevier}
}

License

This program is available for download for non-commercial use, licensed under the GNU General Public License, which is allows its use for research purposes or other free software projects but does not allow its incorporation into any type of commerical software.

Files

The repository contains following files.

• main.py, Python script file, containing the implementation for training and test phases of the GTSNet,
• model.py, Python script file, containing the PyTorch implementation of the GTSNet,
• utils.py, Python script file, containing a collection of small Python functions,
• README.md.