🔍 About | 🔨 Setup | 🚢 Pretrain | ⛵ Finetune | 🚀 Quick Start | 🔗 Citation

🔥 NEWS: Thanks to over 100 stars! We've further refined the code for improved stability. Appreciate your patience as we refine the implementation — ongoing EEG research continues to shape the development of a standardized pipeline.

🔥 NEWS: The paper "CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding" has been accepted by ICLR 2025!

🔍 About

We propose CBraMod, a novel EEG foundation model, for EEG decoding on various clinical and BCI application. The preprint version of our paper is available at arXiv. The camera-ready version of the paper will be available at OpenReview.

🔨 Setup

Install Python.

Install PyTorch.

Install other requirements:

pip install -r requirements.txt

🚢 Pretrain

You can pretrain CBraMod on our pretraining dataset or your custom pretraining dataset using the following code:

python pretrain_main.py

We have released a pretrained checkpoint on Hugginface🤗.

⛵ Finetune

You can finetune CBraMod on our selected downstream datasets using the following code:

python finetune_main.py

🚀 Quick Start

You can fine-tune the pretrained CBraMod on your custom downstream dataset using the following example code:

import torch
import torch.nn as nn
from models.cbramod import CBraMod
from einops.layers.torch import Rearrange

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = CBraMod().to(device)
model.load_state_dict(torch.load('pretrained_weights/pretrained_weights.pth', map_location=device))
model.proj_out = nn.Identity()
classifier = nn.Sequential(
  Rearrange('b c s p -> b (c s p)'),
  nn.Linear(22*4*200, 4*200),
  nn.ELU(),
  nn.Dropout(0.1),
  nn.Linear(4 * 200, 200),
  nn.ELU(),
  nn.Dropout(0.1),
  nn.Linear(200, 4),
).to(device)

# mock_eeg.shape = (batch_size, num_of_channels, time_segments, points_per_patch)
mock_eeg = torch.randn((8, 22, 4, 200)).to(device)

# logits.shape = (batch_size, num_of_classes)
logits = classifier(model(mock_eeg))

🔗 Citation

If you're using this repository in your research or applications, please cite using the following BibTeX:

@inproceedings{wang2025cbramod,
    title={{CB}raMod: A Criss-Cross Brain Foundation Model for {EEG} Decoding},
    author={Jiquan Wang and Sha Zhao and Zhiling Luo and Yangxuan Zhou and Haiteng Jiang and Shijian Li and Tao Li and Gang Pan},
    booktitle={The Thirteenth International Conference on Learning Representations},
    year={2025},
    url={https://openreview.net/forum?id=NPNUHgHF2w}
}

⭐ Star History