Decoding Nature Images from EEG for Object Recognition [ICLR2024]
Core idea: basic constrastive learning for image and EEG. Interesting analysis from neuroscience perspective! ๐คฃ
- Propose a self-supervised framework for EEG-based object recognition with contrastive learning, achieving remarkable zero-shot performance on large and rich datasets.
- Demonstrate the feasibility of investigating image information from EEG signals, by resolving brain activity from temporal, spatial, spectral, and semantic aspects.
- Apply two plug-and-play modules to capture spatial correlations among EEG channels, offering evidence that the model discerns the spatial dynamics of object recognition.
many thanks for sharing good datasets!
- Things-EEG2
- Things-MEG (updating)
./preprocessing/
- raw data:
./Data/Things-EEG2/Raw_data/
- proprocessed eeg data:
./Data/Things-EEG2/Preprocessed_data_250Hz/
-
pre-processing EEG data of each subject
- modify
preprocessing_utils.py
as you need.- choose channels
- epoching
- baseline correction
- resample to 250 Hz
- sort by condition
- Multivariate Noise Normalization (z-socre is also ok)
python preprocessing.py
for each subject.
- modify
-
get the center images of each test condition (for testing, contrast with EEG features)
- get images from original Things dataset but discard the images used in EEG test sessions.
Now we release the image features extracted with CLIP model in ./dnn_feature/
.
./nice_stand.py
./visualization/
Hope this code is helpful. I would appreciate you citing us in your paper. ๐
@misc{song2023decoding,
title = {Decoding {{Natural Images}} from {{EEG}} for {{Object Recognition}}},
author = {Song, Yonghao and Liu, Bingchuan and Li, Xiang and Shi, Nanlin and Wang, Yijun and Gao, Xiaorong},
year = {2023},
month = nov,
number = {arXiv:2308.13234},
eprint = {2308.13234},
primaryclass = {cs, eess, q-bio},
publisher = {{arXiv}},
doi = {10.48550/arXiv.2308.13234},
archiveprefix = {arxiv}
}