Open EEG Bench ships with the following pretrained EEG foundation models:
| Backbone | Reference | Pretrained Source | Normalization |
|---|---|---|---|
| BIOT | Yang et al. 2023 | braindecode/biot-pretrained-prest-16chs |
Percentile scale (q=95) |
| LaBraM | Jiang et al. 2024 | braindecode/labram-pretrained |
Divide by 100 |
| BENDR | Kostas et al. 2021 | braindecode/braindecode-bendr |
Min-max scale [-1, 1] |
| CBraMod | Wang et al. 2025 | braindecode/cbramod-pretrained |
Divide by 100 |
| SignalJEPA | Guetschel et al. 2024 | braindecode/SignalJEPA-pretrained |
None |
| REVE | Music et al. 2025 | brain-bzh/reve-base |
Z-score (clip σ=15) |
These are available as factory functions in open_eeg_bench.default_configs.backbones:
import open_eeg_bench as oeb
oeb.default_configs.backbones.biot()
oeb.default_configs.backbones.labram()
# etc.