Figure 3.

Framework of MViT for multi-channel EEG feature learning. It consists of a stack of N transformer encoders; each encoder processes image tokens from an individual EEG channel. The output feature representations are then concatenated and fed as an input to MLP for EEG classification.