PierreGtch/EEGNetv4

Model Card for pre-trained EEGNet models on mental imagery datasets

Collection of 12 neural networks trained for motor imagery decoding along with evaluation results.

Model Details

• Architecture: EEGNetv4 by Lawhern et. al (2018).

How to Get Started with the Model

• Download and load in memory:

import pickle

# download the model from the hub:
path_kwargs = hf_hub_download(
    repo_id='PierreGtch/EEGNetv4',
    filename='EEGNetv4_Lee2019_MI/kwargs.pkl',
)
path_params = hf_hub_download(
    repo_id='PierreGtch/EEGNetv4',
    filename='EEGNetv4_Lee2019_MI/model-params.pkl',
)
with open(path_kwargs, 'rb') as f:
    kwargs = pickle.load(f)
module_cls = kwargs['module_cls']
module_kwargs = kwargs['module_kwargs']

# load the model with pre-trained weights:
torch_module = module_cls(**module_kwargs)

• Details: more details and potential use-case scenarios can be found in the notebook here

Training Details

• Training dataset: Each model was trained on the dataset with corresponding name in the MOABB library (see datasets list).
• Details: For details on the training procedure, please refer to the poster here.

Evaluation

• Cross-dataset transfer: The transfer abilities of the models was tested on the same datasets as for training.
• Details: The evaluation procedure can be found in the poster here and the article Transfer Learning between Motor Imagery datasets using Deep Learning.
• Results: The evaluation results can be found under the results/ folder.

Model Card Authors

• Modedels training and results by: Pierre Guetschel