ycq091044/BIOT

Normalizes heterogeneous EEG data—varying channel counts, lengths, and missing values—into consistent token sequences processed by a transformer encoder, enabling cross-dataset pretraining. Provides multiple pretrained checkpoints (5M+ samples from resting EEG, sleep EEG, and seizure datasets) that transfer effectively to downstream tasks like abnormality detection. Includes baseline implementations of SPaRCNet, ContraWR, and ST-Transformer alongside supervised and unsupervised pretraining pipelines.

182 stars. No commits in the last 6 months.