martinwimpff/eeg-otta

Calibration-free online test-time adaptation for electroencephalography motor imagery decoding

This is the official repository to the paper Calibration-free online test-time adaptation for electroencephalography motor imagery decoding. The implementation is based on mariodoebler/test-time-adaptation. Additionally we use BaseNet from this repository.

Usage

Installation

• clone this repository
• run pip install . to install the eeg-otta package

Note: you can also use poetry for the installation

Source training

• run train_source_model.py with the --config of your choice, the checkpoints and the config will automatically saved in the checkpoints directory

Note: you can also use one of the checkpoints in the checkpoints directory

Run the online test-time adaptation

• run run_adaptation.py with the --config and source_run of your choice (one of the configs starting with tta)
• the setting (cross-session or cross-subject/ cross-subject continual) is dependent on your checkpoint i.e. whether the within-subject dataset (_within) or the leave-one-subject-out (_loso) dataset was used.
• To choose between the cross-subject and cross-subject continual setting, modify the continual parameter in the TTA config file (cross-subject is the default).

Citation

If you find this repository useful, please cite our work

@inproceedings{wimpff2024calibration,
  title={Calibration-free online test-time adaptation for electroencephalography motor imagery decoding},
  author={Wimpff, Martin and D{\"o}bler, Mario and Yang, Bin},
  booktitle={2024 12th International Winter Conference on Brain-Computer Interface (BCI)},
  pages={1--6},
  year={2024},
  organization={IEEE}
}