OneFi: One-Shot Recognition for Unseen Gesture with COTS WiFi

This repo contains the code of the following paper:

OneFi: One-shot Recognition for Unseen Gesture with COTS WiFi

Tested Environment

We tested our codes under this environment:

• MATLAB 2020b
• Python 3.8.5
• PyTorch 1.7.1

How to use the code

• Please follow the instructions under virtual-gesture-generation folder to perform data augmentation. (corresponding to Sec.4 in paper).

  • Add the virtual-gesture-generation folder to the path in MATLAB first.
  • Run compute_velocity_distribution.m.
    • This function will take raw CSI data (.dat file) as input and output corresponding velocity distribution.
    • The format of computed velocity_distribution is of shape: [velocity_x_bin, velocity_y_bin, timestamps].
    • Save the velocity_distribution as .mat file for further usage. Code example: save("1-1-1-1.mat", 'vd').
    • This function would take around 30 mins to run because of the inherent slow optimization computation.
  • Run create_virtual_gesture.m.
    • This function will take the computed virtual gesture as input from the .mat file.
    • The input parameter direction is the rotation angle for data augmentation. In our paper, we set direction as (30, 60, … 330) so that our base dataset is expanded by 12 times (Please refer to our paper for more details.)

• Please follow the instructions under few-shot-learning folder to perform one-shot learning (corresponding to Sec.5 in paper). Please replace train_data.npy and test_data.npy with your own data.

• Please feel free to open an issue or send us an email if there are any questions about the paper and the code usage.

Citation

If it is useful for your research, please consider cite the following reference paper:

@inproceedings{onefi,
  author    = {Rui Xiao and
               Jianwei Liu and
               Jinsong Han and
               Kui Ren},
  title     = {OneFi: One-Shot Recognition for Unseen Gesture via {COTS} WiFi},
  booktitle = {SenSys '21: The 19th {ACM} Conference on Embedded Networked Sensor
               Systems, Coimbra, Portugal, November 15 - 17, 2021},
  pages     = {206--219},
  publisher = {{ACM}},
  year      = {2021},
  url       = {https://doi.org/10.1145/3485730.3485936},
  doi       = {10.1145/3485730.3485936},
  timestamp = {Mon, 15 Nov 2021 13:08:25 +0100}
}

Appendix

Training Set:

Testing Set:

H: horizontally; V: Vertically

CW: clockwise; CCW: counterclockwise