Data is here, pre-trained model on action detection is here, and pre-trained model on action classification is here.
- train_data_amp: CSI amplitude of training data: 1116x52x192, 1116 CSI series, each series has 52 carriers, and 192 samples
- train_label_instance: action labels, 1116x192, 1116 CSI series, each series and 192 samples that are labeld from 0 to 6 (background + 6 actions).
- train_label_mask: action labels, 1116x192, 1116 CSI series, each series has 192 samples that are binaryly labeled.
- train_label_time: start time and end time labels. 1116x2, 2 is for start and end.
similar to the above
Temporal operations including 1d convolutions, 1D maxpoolings, and 1D deconvolutions sweep along the time axis of CSI series.
If this helps in your research, please kindly cite,
@article{wang2019temporal,
title={Temporal Unet: Sample Level Human Action Recognition using WiFi},
author={Wang, Fei and Song, Yunpeng and Zhang, Jimuyang and Han, Jinsong and Huang, Dong},
journal={arXiv preprint arXiv:1904.11953},
year={2019}
}