The test accuracy is only around 20%.

[Qingfengxulailu]

I download WiGesture data，https://huggingface.co/datasets/RS2002/WiGesture/tree/main
unzip WiGesture.zip, run process1.py, process2-split.py,
run cross_class-one_shot.py， finally， the accuracy on the test set can only reach about 20%.

[RS2002]

Hi!

Thank you for your interest in our work.

The data processing code was originally written for the CSI-BERT project, where we mark lost packets as [PAD] (set to -1000). Before using the dataset in other projects, you may need to interpolate the [PAD] values first (in the CrossFi paper, we use the linear interploation).

Sorry for the lack of clarity on this point (several people have also contacted us via email with the same question). We will update the README soon. I hope this helps.

Best Regards,
Zijian