U-Shape Networks are Unified Backbones for Human Action Understanding from Wi-Fi Signals

Prerequisite

• numpy
• pandas
• scipy
• torch
• tqdm
• scikit-learn

How to run

0. Download the dataset ARIL from its project.

   Download the dataset WiAR from its project.

   Download the dataset HTHI from here

1. "git clone" this repository.

2. Datasets ARIL and HTHI do not require processing. Datasets ARIL and HTHI do not require processing，

   1. unzip WiAR dataset and cd create_wiar_dataset
   2. run python load_data.py to get csi_amp_all.mat
   3. run python traintestsplit.py <index> (index is an int type, indicating the round of random division)
   4. get TestDataset1.mat and TrainDataset1.mat

3. Run bash run.sh (If you want to run Gaussian mode detection, please 'bash run_detection_gaussian.sh')

Input Parameters:

python train_eval.py --model_name <model_name> --task <task> --dataset_name <dataset_name>

• --model_name: choose between unet, unetpp and fcn
• --task: choose between classify, detection, and segment
• --dataset_name: choose between HTHI, WiAR and ARIL

Please note that when the dataset_name is set to HTHI, the task parameter can only be set to detection.

gaussian smooth label

run gaussian_smooth_label.py

Citation

If this helps your research, please cite our paper.

@article{wang2023wifiushape,
 title={U-Shape Networks are Unified Backbones for Human Action Understanding from Wi-Fi Signals},
 author={Wang, Fei and Gao, Yiao and Lan, Bo and Ding, Han and Shi, Jingang and Han, Jinsong},
 journal={IEEE Internet of Things Journal},
 year={2023},
 publisher={IEEE}
 }