EfficientFi Human Activity Sensing (HAS)

CSI-based WiFi sensing benchmark used to evaluate WavesFM on human activity recognition.

Official links

• Raw data: https://github.com/xyanchen/WiFi-CSI-Sensing-Benchmark
• Paper: EfficientFi: Toward Large-Scale Lightweight WiFi Sensing via CSI Compression (IEEE IoT Journal 2022)

Tasks supported

• Human activity recognition (Mean Per-Class Accuracy)

Class labels

• run
• walk
• fall
• box
• circle
• clean

Split

• 80/20 random split

Preprocessing

1. Load CSIamp from each .mat file and reshape to (3, 114, T).
2. Cast to float32 tensor.
3. Resize to 224x224 (bicubic, antialias).
4. Min-max normalize with dataset-wide bounds (2.44 to 54.72).
5. Standardize per channel with dataset-wide mean/std.
6. Parse label from filename prefix (run, walk, fall, box, circle, clean) and write to HDF5.

Script: preprocess_csi_sensing.py

python preprocessing/preprocess_csi_sensing.py \
  --data-path <HAS_DIR> \
  --output data/has.h5

Metric

• Mean Per-Class Accuracy on the test split.

Citation

@ARTICLE{9667414,
  author={Yang, Jianfei and Chen, Xinyan and Zou, Han and Wang, Dazhuo and Xu, Qianwen and Xie, Lihua},
  journal={IEEE Internet of Things Journal},
  title={EfficientFi: Toward Large-Scale Lightweight WiFi Sensing via CSI Compression},
  year={2022},
  volume={9},
  number={15},
  pages={13086-13095},
  doi={10.1109/JIOT.2021.3139958}
}